The present invention relates to a method and a system for determining the velocity of a moving fluid surface by means of at least one camera .

Measuring the velocity of a flow is needed in many

applications , e.g. when determining the streamflow in an open channel . Such measurements are for instance important in the realm of irrigation, drinking water supply,

hydroelectric power production, flood control , reservoir control , sewage systems , preservation of ecosystems , etc . Streamflows can occur in many different types of structures , natural courses , artificial channels , furrows etc . All of them are open-channel flows , i.e. flows having a free surface .

There are many different systems available for measuring the velocity . Systems which are image based possess the

advantage that there is no need of expensive installations and they offer more flexibility than other non- intrusive measurements systems . A long established technique for velocity measurements is the technique of Particle Image Velocimetry (PIV) ; see e.g. R.J. Adrian, 1991 , " Particle- imaging techniques for

experimental fluid mechanics . " Annual Review of

Fluid Mechanics 23, 261-304. PIV is since the work of Ichiro Fuj ita, Marian Muste , and Anton Kruger (1998 , "Large-scale particle image velocimetry for flow analysis i hydraulic engineering applications . " Journal of Hydraulic Research 36 (3) : 397-414) also known of being applied to large scale free surface flows of flumes or open channels . This flavor of PIV is known to the hydraulic research and engineering community as Large Scale PIV (LSPIV) , see e.g. - Muste, M., I . Fujita, and A. Hauet . 2008. "Large-Scale Particle Image Velocimetry for Measurements in Riverine Environments . " Water Resources Research 44 (4 ) , and

- Muste, M. , H.-C. Ho, and D. Kim. 2011. "Considerations on Direct Stream Flow Measurements Using Video Imagery:

Outlook and Research Needs." Journal of Hydro-Environment Research 5 (4): 289-300.

Recent exam les of LSPIV being applied to river flows are described e.g. in

- patent application WO 2014/013064 Al,

- Kim, Y . , M. Muste , A. Hauet , W . F. Kraj ewski , A. Kruger, and A . Bradley. 2008. "Stream Discharge Using Mobile Large-Scale Particle Image Velocimetry : A Proof of

Concept . " Water Resources Research 44 (9),

- Dramais , Guillaume , Jerome Le Coz , Benoit Camenen, and Alexandre Hauet . 2011. "Advantages of a Mobile LSPIV Method for Measuring Flood Discharges and Improving

Stage-discharge Curves . " Journal of Hydro-Environment Research 5 (4) : 301-12 , and

- Tsubaki , Ryota , Ichiro Fuj ita , and Shiho Tsutsumi . 2011.

"Measurement of the Flood Discharge of a Small-Sized River Using an Existing Digital Video Recording System . " Journal of Hydro-Environment Research 5 (4): 313-21.

All these mentioned PIV methods have in common that they require having a well detectable flow tracer of some kind . However, the use of natural or artificial tracers possesses some complications to practically measure the velocity , since the tracers are not generally present and/or cannot be added permanently and continuously .

In the method disclosed in WO 2014/013064 Al the velocity of the water surface is determined by subtracting two images to obtain a composite image . Subsequently, a suitable threshold is chosen to suppress the non-moving zones and a PIV

analysis is performed to determine the velocity of some tracers . Since the displacement of the tracers is determined within the same composite image, their direction of movement is indefinite.

The depth of water can also be estimated using images . In WO 201.4/013064 Al it is proposed to analyze pixel colorimeter to determine the water level. However, using pixel

colorimeter can possess some difficulties for analyzing images with poor light conditions . The method described in WO 2014/013064 Al also needs at least one visual reference object being partially immersed in water, which constrains the places where the system can be installed . In addition, at least 6 reference points which have to be geo- referenced are needed for the camera calibration, in which the external parameters (i.e. position and orientation and at least the internal parameter focal distance) are obtained.

It is an aim of the present invention to provide for a method and a system which allow determining in a reliable manner the velocity of a mo ing fluid surface without need of adding flow tracers .

This aim is achieved by the method and the system as defined in claims 1 and 13 , respectively. The further claims specify preferred embodiments of the device and the system according to the invention as well as computer program and a data medium .

The invention is explained in the following by means of exemplary embodiments with reference to Figures . In the drawings :

Fig . 1 is a flowchart of steps of a method according to an embodiment of the invention; Fig . 2 shows a schematic view of a first embodiment of a system for performing the method illustrated in Fig . 1 ; Fig . 3 shows a schematic view of a second embodiment of a system for performing the method illustrated in Fig . 1 ;

Fig . 4 shows a schematic view of a third embodiment of a system for performing the method illustrated in Fig . 1 ;

Fig . 5 shows a schematic view of the relationship between world and camera coordinates ; Fig . 6 shows a schematic view of an embodiment for calibrating a camera ;

Fig . 7 shows a schematic view of another embodiment for calibrating a camera ;

Fig . 8 is a flowchart of the sub-steps of step 500 of the method of Fig. 1, wherein in the images acquired "wet " and "dry" parts of the scene are distinguished from each other ; Fig . 9 is a flowchart of the sub- steps of step 600 of the method of Fig . 1, wherein the water level is determined based on the images acquired;

Fig. 10 is a schematic illustration of the method for determining the water level for the embodiment of Fig . 3;

Fig . 11 is a flowchart of the sub- steps of step 700 of the method of Fig . 1, wherein the surface flow velocity is determined;

Fig . 12 is a flowchart of the sub- steps of step 800 of the method of Fig . 1, wherein the discharge is determined, and

Fig . 13a 13c refer to an example of three images, which are processed according to the method step 700 of Fig . 1.

The following described embodiments concern a method and a system for obtaining the surface velocity field and, if required, the fluid level in an open channel by means of one or more digital cameras . The measured quantities may be combined with a priori knowledge on the geometry of the channel to determine e.g. the discharge or other quantities . The embodiments are described with respect to flowing water, but they are applicable also for other flowing fluids .

An open channel is a conduit with a free surface , i.e. the fluid flowing in the channel has a free surface (e.g. river, stream, artificial channel , furrow, etc . ) . Streamflow or discharge, Q, is defined as the volumetric rate of flow of the fluid in the channel and is given for instance in units of cubic meter per second. A priori knowledge on the

geometry of the channel may include e.g. information on the profile in the spanwise direction of the channel . The information may be e.g. the height H above the surface defining the channel in function of y, where y is the distance in the spanwise direction of the channel , see Fig . 2. In a simple case , where the channel can be considered as having a rectangular profile , information on the distance between the side walls limiting the channel and its depth may be sufficien . Similarly, a channel having a parabolic or trapezoidal profile can be parameterized with only a few, e.g. two, parameters .

The method for determining the velocity and, if required, the streamflow in an open channel comprises the following steps , see Fig . 1: - Step 200 : Set-up of the hardware, in particular

installation of at least one camera. The camera may be any imaging device which is suitable to provide for a sequence of digital images . Apart from usual digital cameras, it is conceivable to use mobile phones , such as s artphones , or webcameras .

Step 300 : Calibration of the camera ( s) .

- Step 400 : Image acquisition . - Step 500: Processing of the images acquired in order to distinct between "wet" and "dry" parts of the scene.

- Step 600 : Water level determination based on the images acquired .

- Step 700 : Image processing to determine the surface flow velocity .

- Step 800 : Streamflow calculation .

In the following the steps 200-800 are explained in more details in relation to three camera configurations CI, C2 , C3 :

-- Camera configuration CI: usage of a single camera 203 , which is fixed, see Fig . 2,

-- Camera configuration C2 : usage of two cameras 203 and

203a, which are fixed and are arranged for a stereo gauging, see Fig . 3.

- Camera configuration C3 : usage of a mobile device 203b, e.g. a mobile phone , hold by a user, see Fig. 4.

Step 200: General set-up

Fig . 2 shows schematically an open channel , which is defined by the boundary surface 201. The water flowing in the channel defines a water line 202a, which is given by the intersection of the water surface 202 with the boundary surface 201 of the channel. As the channel is open, the water surface 202 is also denoted as a free surface . The arrows 202b illustrate the velocity of the water surface 202 at different locations .

In the camera configuration CI one camera 203 is installed at a place , where it can view a section of the water surface 202 , which is moving, and a section, which is non-moving, such as a wall 201a being a part of the boundary surface 201 above the water line 200a or another non-moving obj ect being in contact with the water surface 202. The camera 203 does not need to be placed above the open channel ; it may be placed on one side of the channel , which makes the

installation particularly simple and cheap .

The images acquired by the camera 203 can be stored in a memory 205 , which may be e.g. an internal memory of the camera 203 or an external hard disk . Unit 204 in Fig . 2 is a means for transferring the data to the memory 205 and for powering the camera 203. If the memory 205 is an external one , the unit 204 may be connected to the camera 203 by means of a cable , e.g. an Ethernet cable and may be

configured such that electrical power is passed via this cable to the camera 203 , e.g. i the form of power over Ethernet (PoE) » For processing the images stored, the memory 205 is

connected to processing unit 206 , which may be any type of computer which is configured to perform the necessary calculations . In the configuration C2 shown in Fig . 3 two cameras 203 and 203b are arranged such that by means of stereo image

processing the water level 202a can be determined without making reference to a non-moving obj ect 201a . Thus , the two cameras 203, 203b may be arranged such that they view the moving water surface 202 only . This configuration makes the system particularly flexible .

The cameras 203 and 203b may be geo-referenced. Depending on the width of the channel this is achieved with a temporary mechanical setup for markers with known world coordinates . For larger channels of the order of 100 m, where the width may prohibit such a mechanical setup, for instance a drone or another aerial vehicle may be employed, which can be seen by a camera 203 , 203b, and which can record its position with an accurate DGPS (Differential Global Positioning System) . The cameras 203, 203b are configured such that the velocity is measured for the available field (s) of view . It may or may not cover the entire span-wise direction. In the latter case , a span-wise velocity profile may be fitted to all the available velocity observations . The fitting order depends on the span-wise coverage and the fit obeys the non-slip boundary conditions (i.e. zero velocity) at the channel side walls . The two cameras 203 and 203a are connected to uni 204a, which may e.g. a switch box for transferring the data to memory 205 connected to processing unit 206 and for powering the cameras 203 and 203a . It is conceivable that the memory 205 and/or the processing unit 206 are integral parts of the camera 203 and/or 203a .

In the camera configuration C3 shown in Fig . 4 a user holds one mobile camera 203b, e.g. a smartphone , such that a section of the water surface 202 and a non-moving obj ect , e.g. part of the wall 201a, is imaged at the same time . If the velocity is to be determined along the channel in the whole spanwise direction, the camera 203b is arranged such that it views at least one border of the surface 202.

Preferably, the camera arrangement optimizes the trade of between orthogonal view to the channel wall 201a and between orthogonal view to the water surface 202. This optimum is close to 45 deg inclination to the water surface 200. If need be , a spanwise velocity profile is fitted to all the available velocity observations to get the velocity data in the whole spanwise direction .

Step 300 : Calibration

For determining the desired quantities (flow velocity, discharge , etc . ) a camera model is used which describes the mathematical relationship between the coordinates of a point in three-dimensional space ( "world space" ) and its

proj ection onto the image plane . As an example Fig. 5 shows a point P in 3D space , which is defined in the coordinate system 301 by specific world coordinates (x, y, z) , and P ' , which is the point P imaged by a camera on the image plane . The location of P ' in the image plane may be given by two Cartesian coordinates ("image coordinates " ) 302 ( i , j ) . The mapping of the coordinate system 301 on the coordinate system 302 is indicated by arrow 303.

In order to fit a camera model to the camera view, a camera is calibrated, in which the external parameters , i.e.

position and orientation of the camera, and at least one internal parameter, such as the focal distance are obtained . In the following these external and internal parameters are denoted as " calibration parameters " .

One usual possibility for determining the calibration parameters is to provide at least 4 reference points with known world coordinates and to evalua e the points imaged by the camera. For applications , where the scale s of the field of view is large , e.g. s is much larger than 1 m, the world coordinates of the reference points may be determined e.g. by means of DGPS (Differential Global Positioning System) .

As reference points e.g. fixed makers can be used, which are positioned at definite locations at the channel and which are configured to be imaged in a way clearly distinguishable from the background . For this purpose , one or more carriers may be used which are provided with spots having a bright color , e.g. white , and a circular or any other predetermined shape .

In the following various examples of calibration methods are explained with respect to Fig . 2 , 3 , 4, 6 and 7: a) Calibration method using 2 or more reference points :

For one or more fixed cameras, as in the configuration CI and C2 , at least 4 reference points 382 with known world coordinates are provided for, see Fig. 2 and 3 and imaged by the camera (s) . A suitable camera model , e.g. a pinhole model is used, to fit the coordinates of the reference points in 3D space to their coordinates in the image .

In case that there are 4 reference points 382 only, at least one of the following parameters is to be known a priori : the focal distance , the camera ' s vertical distance relative to the reference points 382 , the camera ' s horizontal distance relative to the reference points 382.

In the pinhole model the camera ' s aperture is considered as a point, wherein no lenses are used to focus light . The focal length is the distance between the image plane and the pinhole.

In a similar way at least 4 reference points 382 with known world coordinates may be used to calibrate a mobile camera, see Fig . 4. Preferably , the reference points 382 and the free surface 202 are imaged at the same time . In this way, calibration parameters are available for the images based on which the velocity, water line , etc . are to be determined .

It is conceivable that a priori information of the camera ' s location is available . For instance mobile devices , in particular smartphones , may have a sensor implemented, for instance an accelerometer, which allows a direct measurement of the orientation of the camera 203b relative to the gravity vector 322. In Fig. 6 three perpendicular axes 321 are depicted which define the reference frame 321 of the camera 203b . Its orientation is defined by the angles 321a, 321b, 321c between the gravity vector 322 and the axes of the reference frame 321. By using the information on the camera ' s orientation, two external parameters do not need to be fitted by the camera model . Thus , the calibration method can be simplified, as the focal length, or the camera vertical or horizontal distance relative to the reference points 382 does not anymore need to be known a priori , and a set of 4 reference points 382 is fully sufficient to

calibrate the inhole model,

The methods for calibration explained so far have in common that they yield the camera ' s position relative to a fixed frame , in which the geometry of the channel is defined . The following methods yield the camera position relative to the free surface only. b) Calibration method without usage of reference points

A calibration procedure is possible by using a body which has a known length scale and which is floating on the surface 202 when the images are acquired. This body may be e.g. in form of a dumbbell : A bar of definite length has ends which are provided with floating elements , e.g. hollow spheres .

Step 400 : Image acquisition

In the configuration CI and C2 a camera 203, 203a may be connected e.g. to an Ethernet cable . Energy is supplied to the cameras 203, 203a via PoE . Images are recorded in the memory unit 205 and processed by the processing unit 206.

For determining for instance the water level not only a single image, but a sequence of images are analyzed. The number of images acquired is at least two and may be

typically up to 100 or more . Usual digital cameras can take 30 frames per second, thus the sequence of images may last a few seconds .

The images can be recorded in gray values . Depending on the application, a camera with a moderate pixel resolution may be sufficient , e.g. 640x480 pixels .

During night the scene can be illuminated with an infrared beamer . For allowing a stereo gauging in the configuration C2 the cameras 203 , 203a are arranged such that they take images from the same scene . Preferably, they are also synchronized such that they take images at the same time . However, a time difference of one frame , e.g. 1/30 sec , can be accepted for later analysis .

In the camera configuration C3 the necessary images can be acquired by recording a few seconds of movie , e.g. 5 sec . In this configuration C3 the camera 203b is not fixed . The subsequent procedures for water level determination and surface velocity measurements may require that position and orientation of the recording camera is sufficiently constant over the time of the sequence and sufficiently identical to the position and the orientation of the camera during calibration. For hand-heId smartphone recordings this may not be the case . Image processing can be used to stabilize the image sequence . One possible method for such a

stabilization is as follows :

The came a 203b is hold such that the images acquired contain a significant part that is not water surface 202 , i.e. that contains dry shore scenery 201a and preferably the reference points 382 for calibration or other fixed markers, see Fig . 4 or 6. In the smartphone application described below, the software is configured to guide the user such that the top and bottom quadrants of the images , i.e. the region at the four corners of the image , contain only dry scenery . From these "dry" quadrants the relative movement between individual frames is measured by an image cross- correlation . The cross- correlation can be facilitated by markers fixed at the shore . Preferably, these markers are reference points 382 , which can be also used for

calibration . The resulting motion vectors allow a detection and compensation of the inter- frame movement by image registration ( transformation of the data into one coordinate system) or simpler by a combination of image rotation and translation . Alternatively, and to save CPU power, only those images which are at approximately the same position and orientation as during calibration are used for further processing .

Step 500 : Distinction between "wet" and "dry" image regions

Fig . 8 is a flow chart of the different sub-steps 501 507 of step 500.

Analyzing not only a single image , but a sequence of images allows discriminating between the "wet " region 202 of the image (i.e. flowing water) and the "dry" region 201a, which views non-moving scenery (e.g. the shore , rocks , part of a bridge pillar, etc.) . Any pixels looking to the "dry" parts of the scene will experience little change in their gray values , whereas "wet" parts are subj ect to constant change , caused by little waves and surface perturbations of the flow. In an ideal camera and in an ideal illumination situation the pixel change in the dry region would be zero. However, in reality, there exists a finite noise level due to non- constant electronic camera gain, due to changing light situations (e.g. when clouds are crossing the sky) as we11 as due to mechanical vibrations . The electronic noise increases in dark situations , especially during the night when additional light sources like e.g. infra-red diodes are necessary for the cameras to see anything at all .

Thus , after reading the images out from the memory, step 501 in Fig . 8 , they are filtered to get data im_f , step 502. Noise may be adequately filtered by using for instance a two-dimensional spatial median filter . The kernel width of the filter is chosen to be larger than the noise , which may be in the order of 1 or more pixels , and smaller than the advected patterns on the surface 202 , which may in the order of 10 or more pixels . For instance a kernel of 5 pixel x 5 pixel is conceivable . In order to obtain a scalar measure of the change of each pixel gray value over a defined sequence, the maximum and minimum are determined for each pixel in step 503. This provides image data max ( im_f ) and min ( im_f ) , respectively .

In step 504 , the difference between the maximum and the minimum gray values for a given image sequence is taken to get image data De_im. In quasi-continuous mode , in which data are to be obtained for a longer period of time, a gliding maxima and minima is employed with a kernel width of a given time duration. For instance data are obtained over a period which may last one or more hours , so that the kernel width may be several minutes . A typical kernel width has e.g. a period of 10 minutes. Any extreme event for a pixel will fade out after that the period given by kernel width and it is gradually replaced by its corresponding gliding average value , see step 505.

Depending on the type of scene imaged, additional image processing may be applied to filter out artifacts . For instance the dry region may have relatively sharp

transitions between bright and dark areas , e.g. there can be strong local gradients in gray values . Such gradients may be augmented by pixel noise or mechanical camera vibrations . Examples that may lead to such situations are brick walls , rails , trees and other objects, which are non- completely plane flat . Said transitions may lead to an unwanted signal in the "dry" region in the images De_im, such that the water line cannot be determined correctly . To deal with such artifacts , average image data, Av_ im, are obtained in step 505 by averaging over the unfiltered image sequence , or, in quasi -continuous mode , by determining the unfiltered gliding average over the image sequence with a specific kernel width, which may be of several minutes .

In step 506 , the gradient of Av_ im is obtained, denoted by grad (Av_im) in Fig . 8. In step 507 , this gradient is subtracted from the "max-min difference" , De im, in order to obtain processed image data DG_im, i.e. DC im = De_im - grad (A im) . The distinction between wet and dry regions is enhanced by using the processed image data DG_im, since the artifacts mentioned above are at least partially filtered out .

Step 600 : Water level determination

Fig . 9 is a flow chart of the different sub-steps 601-608 of step 600.

For the camera configuration CI or C3 the water level may be determined as follows : As the geometry of the shore line and the calibration parameters are known, each possible real-world water- level , h, may be mapped into the image space , c (h) , see step 601 in Fig . 9. For plane channel walls this results in straight lines across the image space . However, arbitrary shore line geome ries can equally well be mapped onto general curves through the image space . For the calculation the height is discretized, i.e. the calculation is done for an initial value for the height and then repeated by a consecutive increase in the height, see the rhombus "if Lines < Possible water lines " in Fig . 9.

In step 602 , the sum of gray values of all pixels belonging to a particular c (h) is determined . Preferably, a filter is then applied, see step 603. A suitable filter is for example a Savitzky-Golay filter , which may be of second order and have a kernel width, w, corresponding to the height of the surface ripples. In one example w is 5 mm.

The sum of gray values may be plotted versus h, which results in a one-dimensional relation sum ( c (h) ) < * h, see step 604. In step 605 , the derivative is determined, d [sum (c (h) ) ] /dh, which is negative and extremal , where h corresponds to the water level, see step 607 defining that the maximum slope of the function sum (c (h) ) is to be found. Preferably before this step 607 , the signal sum ( c (h) ) as well as the

derivative d [sum (d (h) ) ] /dh are filtered with a filter width k, see step 606. If k is not larger than the wave-height at the shore- line , the relevant signal is not filtered out and not biased to any direction, i.e. it can be used as a robust indicator for the water-line position .

For the camera configuration C2 a different approach may be used to determine the water level . A stereo gauging may require that the two cameras look to the same water surface section, i.e. that their views overlap , that preferably the cameras are synchronized in time and that a spatial correspondence between the two camera views can be established, i.e. the angles between the two camera axes is not too large , preferably not larger than 45 degrees .

An image A is generated by applying a specific filter . This filter separates the moving surface of the image information from the still image information associated with shades , non-moving objects and channel bottom . The same filter may be used as in steps 702 , 703 described in more detail below.

The filtered image A , which contains the moving part of the surface only , is projected onto several horizontal planes 202 , 202 ' , 202 ' ' , which each represent potential water levels . Each proj ection is compared with its corresponding proj ection from image B , which is the filtered image from the second camera . This is illustrated in Fig . 10 in a simplified way with a scene that contains only two features , a " full" feature Fl and an "empty" feature F2. The

proj ections of each feature overlap only for the proj ection wi h the correct height , h . In Fig . 10 the features Fl and F2 as seen in image A are represented by full and empty circles , Fla and F2a , respectively . The same features in image B are represented by full and empty squares , Fib and F2b, respectively . As can be seen, when projecting Fla, F2a, Fib, F2b onto a plane ha ing an incorrect height , e.g. plane 202 ' , 202 ' ' , the projection of Fla is separated from Fib and the proj ection of F2a is separated from F2b . An overlap of Fla and Fib as well as of F2a and F2b is given when they are proj ected on the plane 202 having the correct height .

In order to quantify the quality of the overlap generally, the correlation between all the proj ected gray values from image A with their corresponding proj ections from image B may be measured . Thus , the comparison between the

proj ections onto the horizontal plane with the correct height yields a maximum correlation and a minimum mean gray value difference between the corresponding pixels . These extreme points thus identify the correct water level . In order to reduce the necessary computer processing time , this approach may be modified by working only on a sub-set of pixels . From image A a sub- set is proj ected onto a potential horizontal plane , and from this plane the set is proj ected onto the virtual image chip of the other camera . Again, the correlation at the positions of the proj ected sub-set from A with the pixels from B is maximal for the correct height of the horizontal plane . Equivalently, the difference of the two point sets is minimal . Working with a sub- set of pixels reduces the processing time by the ratio of the sizes between the sub-set of the image and the entire image .

Step 700 : Surface velocity field Fig . 11 is a flow chart of the different sub-steps 701-705 of step 700. The method as explained in the following allows the surface velocity field to be measured . This method will also be designated as Filtered Delta Image Velocimetry (FDIV) . If the bottom of the channel is visible through the water phase, if shades or scenery reflections are visible on the water surface , and/or if non-moving obj ects like standing waves or solid rocks or bridge pillars "contaminate" the image scene , standard PIV does not work, as it would measure some value between zero and the actual surface velocity.

FDIV is capable to separate in a robust fashion the moving image contents from the still image contents . The

functionality of FDIV does not require the addition of flow tracers .

The image sequence taken by a camera is grouped into image triplets , each containing a first image A, a second image B , and a third image C , see step 701.

For instance at time t _ 1 an image I_l is acquired, at a later time t 2 an image 1 2, at a further later time t 3 an image I_3 , etc . This gives a series of images I I , I_2 , l_3 , etc . The first triplet may be formed for instance by (I_l, I_2 , I_3) , the second triplet by (I_2, I __3 , I_4 ) , the third triplet by ( I_3 , I_4 , I_5 ) , etc . However , it is conceivable that the triplets are formed in a different way : For

instance none of the images of a triplet may be contained in the next triplet, e.g. (I_l, I_2 , I_3) , ( I_4 , I_5 , I_6) , and/or there may be gaps within a triplet , e.g. ( I I , I_3 , I_5 ) . Each triplet will give data on the velocity field for one instant of time . In general , a triplet is formed by ( I_i , I_j , I_k) , wherein i<j <k . In the following processing of one triplet (A, B , C) is explained . In step 702 , the absolute difference between A and B and the absolute difference between B and C is formed giving im_l and im_2, respectively . In step 703 , im_ 1 and im 2 are filtered to produce im_1f and im 2 f . A spatial Gaussian kernel may be e.g. used as a filter. The filter width is chosen large enough so as to remove difference noise in im 1 and im_2 and to distribute or smear small features in their respective proximities from im_ If to im_2f . On the other hand , the filter width is chosen small enough in order not to completely remove the motion signal in im_lf and im_2f . The upper bound for the filter width may correspond to the minimal sub-window size that is defined for the following processing step 704. In order to detect motion displacement of scale d in units of pixels , the sub window scale s in units of pixels is chosen such that it is at least twice d or larger : s --- 2d .

In step 704 , the spatial shift of patterns from im_ If to im 2f is determined .

The images im If and im 2f are divided into sub -windows with size nxm ( " interrogation window" ) . Basically, a su -window of im If is compared with the a sub-window of im_2f in order to find a displacement giving the best match between the two sub-windows . The degree of match may be determined by means of the cross - correlation function of im If and im 2f : In order to calculate the cross- correlation function, the image data are converted to the spectral domain, i.e.

Fourier space by using e.g. a Fast Fourier Transform .

The Fourier transform of im 1f is multiplied with the

Fourier co j ugate transform of im 2f and then the inverse Fourier transform is applied . This gives for each sub-window an impulse -like function for R (x, y) . The location of the impulse relative to a sub-window center relates to the relative pixel displacement between the sub-windows of im_ If and im 2f . The obtained displacement vectors at discrete positions are finally mapped from pixel space to world space by employing a camera model that is obtained with one of the above described calibration methods .

Alternatively , im If and im 2f may first be registered into world space , i.e. the image coordinates are converted to the world coordinates . Subsequently, the ma imum of the cross - correlation function for each sub-window and the

corresponding surface velocity are determined in an

analogous way as explained above . The latter flavor allows one to choose more flexible aspect ratios for the n*m sized sub-windows , as now the sub-window sides can be chosen exactly parallel and orthogonal to the flow direction . On the other hand image registration in world space is more CPU costly than j ust mapping the velocity vectors from pixel space into world space .

Fig . 13a- 13c show an example of the images processed by the step 700 for determining the surface velocity . The three images A, B , C in Fig . 13a show part of a river and of a shore at three consecutive instants of time. Fig . 13b are the two images im_lf and im_2f obtained after applying steps 702 and 703. Processing of these two images according to steps 704 and 705 gives the velocity of the surface at different locations on the river as shown in Fig . 13c . In an alternative method of steps 701 and 702 a group of four different images of the image sequence is taken to determine the difference : im_l = | A-B_l j and im 2= | B 2 -C [ , where B_2 is an image acquired after the image A and preferably also B _1 have been acquired .

Step 800 : Streamflow calculation from water-level and surface velocity Fig . 12 is a flow chart of the different sub-steps 801-805 of step 800.

In order to determine the discharge Q, information on the velocity profile along the vertical direction is needed . Such information may be obtained by using a model , for instance a roughness dependent mixing length model as suggested by R . Absi , "A roughness and time dependent mixing length equation" , Journal of Hydraulic , Coastal and

Environmenta1 Engineering, Japan Society of Civil Engineers , Vol. 62, 2006 , pages 437-446. In this model, the velocity profile along the vertical profile is estimated by

integrating a so-called roughness dependent mixing length model from the channel bottom all the way to the water surface . The mixing length and kinetic energy are modeled as functions of the height above the bottom and they define the local slope of the velocity profile . The initial slope at the channel bottom, i.e. the bottom boundary condition, is a function of the so-called roughness velocity and of the bottom roughness height . In order to meet the water surface , i.e. to satisfy the top boundary condition, the integration is iterated to find the correct roughness velocity .

The water- level and the surface velocity field are combined with the model to obtain vertical velocity profiles and successively, the sought after discharge values Q .

The span-wise direction, y, of the open surface flow is divided into n segments , step 801 in Fig . 12.

In step 802 , for each segment the vertical velocity profile is modeled with the boundary conditions of the free surface velocity v(y) as measured (step 705 ' ) , the no-slip condition at the bottom (i.e. zero velocity) and the bottom roughness height . The average of each velocity profile results in a bulk velocity vb__(y} in step 803 , which multiplied with its corresponding cross sectional area of the segment , yields the discharge Q_s for each span-wise segment, step 804. The sum over all Q_s finally is the discharge Q, step 805.

The bottom roughness height affects both, the water level and the surface velocity of an open surface flow . The present method allows measuring both these quantities independently . For stream sections with sufficiently little change in stream-wise direction , the surface roughness of the bottom may also be determined by the present method itself , rather than "only" be used for the determination of the resulting discharge .

In the following specific applications of the method are described in more detail :

Fixed camera (s) , and pan , tilt , zoom (web-) camera (s) :

In one application, three rigidly mounted webcams measured the water-level and the surface velocity field across the entire river at quasi - con inuous intervals of 2 seconds . At regular intervals of 10 minutes the information has been combined to estimate the discharge . Three cameras were used to better resolve the entire span wise channel width and the channel wall . Alternatively, just one camera may be used, which has a larger field of view and/or which may be pan, tilt and zoom controllable .

By using e.g. a 70 Watt infra-red beamer for illumination, the method worked also during the night . Thus , the method may be configured for an all year round usage , i.e. 24 hours x 7 days , to determine the surface velocity and the

discharge of free surface flows. The addition of flow tracers was not necessary and a particular constructions for camera moun ing over the river was not required . It is enough if the cameras are mounted at the river side and that they can " see" the water surface and the shore - line . For additional flexibility instead of web cameras non- movably arranged, cameras can be used with controllable pan, tilt and zoom positions . As long as each position can be reliably saved and reached again, calibration can be

performed for each position and from here on the method is the same as the above described fixed camera method.

However, the number of cameras may be reduced to only one single camera . Fixed stereo (web-) cameras and pan , tilt, zoom (web-) cameras

With the above described method of stereo -gauging, see Fig . 3, a camera pair can be aligned to view a common section of the open surface to directly determine the water level without the need to see the shore- line .

With cameras that allow controlling pan, tilting and zooming positions , it is possible to successively scan the entire width of the surface flow . Thus , it is possible to measure with just 2 cameras water- level and surface velocity of the entire surface flow, which allows one to determine the discharge independent of the shore- line visibility .

Smartphone implementation

The above methods , preferably combined with a suitable way to either stabilize images or to identify those image triplets which are relatively stable , can be implemented into a smartphone application. The calibration needs to be performed each time, but on the other hand, calibration is also simplified thanks to the use of the accelerometer sensor that is standard even on the cheapest smartphones . In addition, via GPS , also standard on all smartphones , the measurement location is automatically determined and it is thus straight forward to use the same application for any number of measurement locations . For calibration, each channel measuring site will be equipped with n fixed markers of scale , which may be in the order of 1 cm or more . The markers positions relative to each other and relative to the channel geometry need to be measured once . Also the channel geometry needs to be measured once . The computer program for the mobile device ( "app" ) is designed to guide the user towards an approximate position relative to the channel and to the calibration marks for calibration. The actual recording occurs over a time of period which is in the order of 1 sec or more. This is enough to determine water level and water surface velocity, from which the discharge can be determined .

The method and system described so far have various

advantages:

- The measurement does not require the addition of flow

tracers nor the presence of permanent struc ures in the fluid . Measurement is possible also in conditions where non-moving shades , reflections , standing waves , or solid obi ects are a significant part of the image .

- It is sufficient to acquire monochromatic images , e.g. gray images. Such images are better suited for difficult light conditions than color images .

- Measurement is possible even during night by using e.g. an infrared beamer .

- Water level determination is based on the separation of wet and dry parts of the image via analysis of a sequence of images rather than one single image .

The calibration uses a minimum of 4 reference points in the case of a mobile camera with a fixed focal length ; and, if the viewing angle of the camera relative to the horizontal plane is available , only 2 points are needed .

- Measurement can be done also by means of a mobile camera, e.g. a camera in a mobile phone .

- The method and system is applicable in environmental , hydra l ic , hydrological and other areas . From the preceding description, many modifications are available to the skilled person without departing from the scope of the invention, which is defined in the claims . The method for determining the velocity of a moving water surface is applicable for any kind of fluid, not only water.

Depending on the application, the method may be performed without the steps for determining the fluid level and/or discharge.

Illumination by means of a lightening device , e.g. an infrared beamer is generally conceivable in cases, where the environmental light is insufficient , e.g. also in an closed environment such as a sewage system .