FIELD OF INVENTION

The present invention relates to a method of risk model development, more particularly relates to a method of developing a risk model and using said model to determine safe zones of a system.

BACKGROUND OF INVENTION

Complex behaviors are difficult to detect and this is one of the key issues in current surveillance applications. It is a challenge to detect a subject's behavior that has been recorded from a surveillance application. Using a simple solution based on hit or miss method, a subject is considered to be behaving suspiciously as long as said subject starts deviating from the main or the reference path. This method does not provide satisfactory results as a subject may not necessarily be a suspect as this method will possibly create a significant amount of false alarms.

Subject behavior databases may usually be procured using standard machine learning or any artificial intelligence methods. However, it is the efficient utilization of this data that will guarantee accuracy of prediction.

U.S. 6441734 describes a security monitoring system that records and compares trajectory information of an individual's path against known trajectories. However, the downfall of this method is that its comparing capabilities are able to only produce binary results where an individuals' path is deduced to be suspicious or non-suspicious. U.S. 7127083 describes yet another video surveillance system that uses rule-based reasoning and multiple-hypothesis scoring to detect predefined behaviors based on an individual's movement. Unfortunately, this system is based on the assumption that the hypothesis made is accurate where a faulty assumption such as criteria of size may skew the results of the detection.

Therefore, there exists a need in the field to produce a reliable method of detection and determining a subjects' complex behavior from a surveillance system.

SUMMARY OF INVENTION

Accordingly there is provided a method of risk model development, the method includes the steps of extracting a line profile from an occupancy map, detecting peaks in the line profile, modeling the peak inline with Gaussian distributing equation, populating an occupancy graph by rows and columns, calculating a path deviation risk table and generating a path deviation risk map from the risk table.

There is also provided a method of determining safe zones from a risk model, the model includes the steps of detecting tracking data from input devices, comparing data with a risk model in a provided database and verifying tracked data is within a predefined safety zone.

There is further provided a method of determining safe zones from a risk model, the method includes the steps of detecting tracking data from input devices, comparing data with a risk model in a provided database, retrieving confidence levels of tracking data from the risk model, comparing confidence levels with predefined danger levels, triggering an alarm if danger levels are higher than predefined danger levels, comparing confidence levels with predefined warning levels and alerting security if warning levels are higher than the predefined warning levels.

The present invention consists of several novel features and a combination of parts hereinafter fully described and illustrated in the accompanying description and drawings, it being understood that various changes in the details may be made without departing from the scope of the invention or sacrificing any of the advantages of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

The present invention will be fully understood from the detailed description given herein below and the accompanying drawings which are given by way of illustration only, and thus are not limitative of the present invention, wherein:

Figure 1 is a flowchart showing the steps taken in developing a risk model;

Figure 2 is a diagram illustrating a contour plot of a multiple Gaussian distribution and its corresponding cross-sectional view;

Figure 3 is a diagram illustrating a 3 dimensional view of a surface map with multiple Gaussian distributions;

Figure 4 is a diagram illustrating an original Gaussian distribution and a converted Gaussian distribution and their respective standard deviation graphs;

Figure 5 is a diagram illustrating an effect of changing standard deviation of a Gaussian distribution;

Figure 6 is a diagram illustrating a 3 dimensional view of a final path deviation risk map with an extended risk distribution in an embodiment of the present invention;

Figure 7 is a flowchart showing the steps taken in determining safe zones using a path deviation risk map;

Figure 8 is a diagram illustrating zone distribution for safe levels, warning levels and danger levels in an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

The present invention relates to a method of developing a risk model and using said model to determine safe zones of a system. A detailed description of preferred embodiments of the invention is disclosed herein. It should be understood, however, that the disclosed preferred embodiments are merely exemplary of the invention, which may be embodied in various forms. Therefore, the details disclosed herein are not to be interpreted as limiting, but merely as the basis for the claims and for teaching one skilled in the art of the invention.

The following detailed description of the preferred embodiment will now be described in accordance with the attached drawings, either individually or in combination.

While the following description may describe example embodiments of the present invention in relation to surveillance systems, the invention is not limited to and can be applicable to other types of risk model development where similar advantages may be obtained. Such risk models for which inventive embodiments may be applicable specifically include, financial and medical applications.

Occupancy maps are obtained from databases wherein these databases are built from standard paths or reference paths of subjects in a scene that are learned through machine learning or any existing artificial intelligence methods. The occupancy map is built by plotting frequency of occurrence against coordinates. In order to produce a reliable database, the occupancy map is built over a long period of time so as to capture a sample that is as close as possible to an actual population that is being sampled. In order to define suspicious and non-suspicious behavior, risk levels are assessed to produce a reliable model since the analysis of human behavior is a complex exercise. This invention discloses a method of developing a risk model to assess suspicious behavior of subjects in a surveillance system.

Figure 1 illustrates the steps of developing a risk model. The occupancy map produced from long term learning through a machine learning technique such as a neural network is used as a starting point for this analysis. The occupancy map can be described as a multiple Gaussian distribution on a 2-dimensional surface map. The contour plot of the multiple Gaussian distribution and its corresponding cross-sectional view is shown in Figure 2. The method as shown in Figure 1 begins with extracting a line profile from the occupancy map. The peaks in said line profile are detected. The cross-sectional view of the Gaussian distribution shows a line profile according to the location in the map with respect to the scene. The peaks from the line profile are then modeled in Gaussian and standard deviation is calculated for each point of the entire cross-sectional Gaussian distributions.

All Gaussian distributions are first identified and stored in memory. Standard deviation values for each of the Gaussian distributions are identified and the distribution is then transformed using the formula below:

Where μ = mean and σ = standard deviation. The Gaussian distributions are then used to populate an occupancy graph. Some line profiles in the Gaussian distribution have more than 1 peak. When this occurs, the peaks are placed in the occupancy graph based on the order of which they appear in said Gaussian distributions as seen in Table 1.

Table 1

These steps are repeated until all line profiles are extracted and the Gaussian distribution is used to calculate a two dimensional path deviation risk table as shown in Table 2.

Table 2

This risk table is a numerical representation of the original occupancy map. A cross section of the Gaussian distribution shows a frequency profile according to the location in a map (as seen in Figure 2) with respect to the scene. Figure 3 shows a 3 dimensional view of a surface map where X axis and Y axis denote a coordinate of a plot and Z axis denotes number of occurrences at that particular point.

In order to use a normal distribution model more efficiently, it is deduced that the modeling of the normal distribution against the standard path cannot accurately represent the true scenario of subjects' path in the scene. It is disclosed in this invention that the normal distribution built from an occupancy map is modified in order to produce confidence level information when suspicious behavior is detected. An empty path deviation risk graph is created. A conversion procedure is conducted on every cell in the original occupancy map. The conversion is described as multiplying the standard deviation in the original occupancy map by a predetermined factor, such as 3. The resulting value, which is a new standard deviation value, is replaced into the original equation wherein the newly created Gaussian equation is used to populate the created empty path deviation risk graph as shown in Table 3.

Table 3 Therefore, the new equation features the following condition:

1 σ in the new distribution = 3 σ opening in the original distribution

where σ = standard deviation value.

With this condition applied to the Gaussian equation, 99.54 % of the original distribution from the original equation now uses 1 standard deviation (σ) of the new path deviation risk graph. Figure 4 shows both the original and new Gaussian distributions and their respective σ distribution graphs. In order to produce a meaningful detection mechanism, the original 3 σ distribution is matched to the new 1 σ distribution in the newly created Gaussian distribution. Doing this would open up the dispersion of the distribution. This creates an extended continuous risk distribution as the newly created Gaussian distribution. This is to ensure that we do not obtain a binary result such as a yes or no answer. An effect of changing the standard deviation to a value of 3 σ from the original standard deviation of Gaussian distribution of 1 σ can be seen in Figure 5. A 3 dimensional view of the final path deviation risk map with the extended risk distribution that has been generated from the completed risk table of Table 3 can be seen in Figure 6.

The new final path deviation risk map can be used to detect suspicious and non- suspicious behavior in terms of path deviation in a system. For example, in this embodiment of the invention, the risk map created can be used in a surveillance system. Figure 7 depicts a method of determining safe zones using the path deviation risk map. The surveillance system detects tracking information and compares this information with the risk model in a previously stored database. If the tracked information is within a predefined safe zone, the process ends at this point as it is verified as within a safe zone. However, if the tracked information is not within the safe zone, confidence level of the tracked object will be retrieved from the risk model. This confidence levels are then compared with the predefined danger levels. If the risk level is found to be insignificant, the tracked object will then be compared with predefined warning levels instead. If the warning level is found to be insignificant as well, the tracked object data will be passed back to main detection mode. However, if the warning levels are found to be sufficiently above the predefined levels, security personnel will be alerted of the warning levels. Similarly, if danger levels are found to be sufficiently above the predefined levels, an alarm will be sounded to signify the danger levels. Figure 8 illustrates zone distribution for safe levels, warning levels and danger levels as described. The action that is assigned as a warning level and danger level can be determined by user and the actions described here are not limitative as these actions can be defined by users as required.