The present invention concerns a method and system to assess abnormal driving behaviour of vehicles travelling on a road and aims to improve driving safety.

The invention relates to a driver assistance system that monitors the vehicle's motion and its traffic environment, detects and tracks other vehicles travelling on the road to quantify abnormal driving behaviour of vehicles by information processing from onboard sensors for assisting the driver in safety improvement. The vehicle that implements the invention is further called a host vehicle, and a vehicle travelling in the proximity of the host vehicle is called a neighbouring vehicle.

It is a constant necessity to improve the safety on road and to prevent accidents. Nowadays, vehicles, such as personal cars, are equipped with various sensors which allow characterizing the vehicle's position and motion. Further, some vehicles are also equipped with additional sensors which provide information about other vehicles travelling on the same road, which are called neighbouring vehicles. Such additional sensors include lidars, radars and cameras.

The challenge of preventing road accidents is to detect a volatile behaviour of travelling vehicles, which is indicative of an abnormal driving and an increased risk of road accidents. The detection of abnormal driving can be achieved by: (a) evaluating the own driver state and behaviour along with the vehicle motion parameters, and (b) detecting the abnormal driving of neighbouring vehicles, for example, their meandering or exceeding of the speed limit. In other words, the detection of abnormal driving requires self-awareness and awareness toward others.

The driving behaviour of a travelling vehicle is defined by such characteristics as the path travelled by the vehicle and the motion parameters of the vehicle. An abnormal driving behaviour can be defined by considering the host vehicle's travelling path and motion parameters alone, or in conjunction with its neighbouring vehicles' travelling paths and motion parameters observed.

A particular element which brings volatility to the traffic flow and may lead to road accidents is the meandering of vehicles on a road.

Some known methods use an on-board camera to detect the white-line marking of traffic lanes on the road surface. The images from the camera are then used to calculate the left-hand and right-hand deviations of the vehicle from the lane. However, such methods and systems require on-board cameras and powerful computing means to deal with a large amount of computation for image processing in real time. A driving state monitoring apparatus is proposed in the patent US-B1 -6487575, adapted to calculate a behaviour parameter indicative of the movement and speed of a vehicle. A linear regression of the changes of the behaviour parameter is used to set the regression line as a behaviour reference. A deviation of the vehicle is then computed according to the actual behaviour parameter, the behaviour reference, and the vehicle speed. The quantity of the deviation determines whether the driving state of the driver is proper. The road shape, either straight or curved with a substantially constant radius of curvature, is determined for deciding whether a quantity of lateral movement of the vehicle and, therefore, the driving state of the driver is abnormal. The apparatus proposed in this prior art does not involve an on-board camera, yet it requires substantial calculations, in particular to obtain the road curvature. Additionally, this prior art does not consider neighbouring vehicles travelling on the road, and the linear regression is performed for the yaw angles of the vehicle.

It is an object of the present invention to provide a method and system to assess abnormal driving behaviour of vehicles travelling on a road to improve safety without the use of on-board cameras and image processing calculations, as compared to the prior art.

To this end, the invention proposes a method for assessment of abnormal driving behaviour of a vehicle travelling on a road, using position information and motion information about said vehicle. The method comprises the following steps:

-obtaining and memorizing a plurality of position information items and motion information items, each information item corresponding to a measuring time instant of an observation time interval,

-obtaining at least one estimate of a reference value for said vehicle travelling on said road during the observation time interval,

-calculating a set of deviation values, each deviation value being computed using a memorized information item corresponding to a measuring time instant and the at least one estimate of a reference value,

-obtaining a quantity of deviation from the set of deviation values to quantify a corresponding abnormal driving behaviour,

-repeating the above steps iteratively for monitoring the quantity of deviation for the corresponding abnormal driving behaviour, and

-deciding, based on said quantity of deviation, whether a predetermined threshold for the corresponding abnormal driving behaviour has been exceeded and in case of affirmative result, implementing a measure for safety improvement.

Advantageously, the invention provides a method for calculating a quantity of deviation of the driven host vehicle and/or of its neighbouring vehicles from the corresponding reference values, such as a reference path estimate, a maximum speed allowed on the road or a minimum safe inter-vehicle headway. The calculation is repeatedly carried out on successive time intervals.

According to particular characteristics, the method comprises one or several features as recited in the dependent claims.

According to another aspect, the invention concerns a system for assessment of abnormal driving behaviour of a vehicle travelling on a road, using position information items and motion information items about said vehicle and its neighbouring travelling vehicle or vehicles if any on said road, the system being characterized in that it comprises:

-a processing unit for obtaining a plurality of position information items and motion information items from on-board sensors of said vehicle about said vehicle, each information item corresponding to a measuring time instant of an observation time interval,

-a processing unit for obtaining a plurality of position information items and motion information items from on-board sensors of said vehicle about said neighbouring travelling vehicle or vehicles if any on said road, each information item corresponding to a measuring time instant of an observation time interval,

-a memorizing unit for memorizing a plurality of position information items and motion information items, each information item corresponding to a measuring time instant of an observation time interval,

-a reference values obtaining unit for obtaining at least one estimate of a reference value for said vehicle and said neighbouring vehicle or vehicles if any, travelling on said road during the observation time interval,

-a calculation unit adapted to calculate at least one set of deviation values, each deviation value being computed using a memorized information item corresponding to a measuring time instant and the at least one estimate of a reference value corresponding to the memorized information item,

-an assessment unit adapted to obtain a quantity of deviation for each set of deviation values, and

-a decision making unit, adapted to decide, based on said quantity of deviation, whether a predetermined threshold for the corresponding abnormal driving behaviour has been exceeded to trigger the implementation of a measure for safety improvement.

The invention will be better understood in the light of the detailed description and accompanying drawings listed below, which are only exemplary and by no way limiting:

-Figure 1 represents schematically a host vehicle and its neighbouring vehicle travelling in the same direction on a multiple lane road at the same time instant; -Figure 2 shows a host vehicle and its neighbouring vehicle travelling on a road at a later time instant as compared to the representation of figure 1 ;

-Figure 3 is a functional block diagram of a system according to the invention;

-Figure 4 is a flowchart of meandering calculation for improving driving safety according to a first embodiment of the invention;

-Figure 5 is a flowchart of a meandering matrix calculation according to an embodiment of the invention;

-Figure 6 is a flowchart of a speed excess calculation according to a second embodiment of the invention, and

-Figure 7 represents schematically a host vehicle and its neighbouring vehicle travelling in the same direction on the same lane of a road at the same time instant.

The invention provides a method and system for assessing abnormal driving behaviour to improve driving safety of travelling vehicles, among them a vehicle implementing the method is labelled 2 in figure 1 and said vehicle is also called a host vehicle while a vehicle labelled 10 in figure 1 is called a neighbouring vehicle.

The host vehicle 2 is for example a car, equipped with various sensors, for example an odometer, a steering angle sensor, a lidar, a radar, an inertial sensor, which are not represented in figure 1 . The host vehicle is travelling on a multiple lane road 4, which has lanes respectively labelled 6 and 8 in figure 1 . The host vehicle 2 is travelling on lane 6, and a neighbouring vehicle 10 is running on lane 8, both said vehicles are moving in the same direction.

In the embodiment illustrated in figure 1 , the host vehicle 2 is equipped with sensors which provide information on the neighbouring vehicle 10, travelling in the proximity of vehicle 2, on the lane 8. Only one neighbouring vehicle is depicted in figure 1 , but the invention is not limited to a single neighbouring vehicle of the vehicle 2 which implements the method of the invention. The position of the vehicles is represented at a time instant T _k in figure 1 . The number of traffic lanes depicted in figure 1 and figure 2 is not limiting in either direction of travel.

In an embodiment, the sensors of the host vehicle 2 provide the speed and steering angle of said vehicle 2, which characterize the position of the host vehicle 2 in a known manner. Additionally, the host vehicle 2 may be equipped with a GPS (Global Positioning System), providing the coordinates of the vehicle in a global coordinate system. More generally, a host vehicle 2 is equipped with sensors providing position information items and motion information items of the vehicle itself and its neighbouring vehicles at measuring time instants. Each sensor provides measurements at an approximately regular sampling rate. The host vehicle is also equipped with sensors, such as a lidar or a radar, providing the range distance between the host vehicle 2 and the neighbouring vehicle 10 and allowing for speed calculation of the neighbouring vehicle 10. For example, as shown in figure 2, distance di(T _k ) between the centre of the front end of vehicle 2 and the right- hand back corner of vehicle 10 and distance d ₂ (T _k ) between the centre of the front end of vehicle 2 and the left-hand front corner of vehicle 10 are provided.

The sampling period of the system according to the present method is denoted as τ in relation to position and speed information about the travelling vehicles, for example τ=50 ms.

Figure 1 shows the position of two vehicles, respectively host vehicle 2 and its neighbouring vehicle 10 at time instant T _k . Figure 2 illustrates the position of the same two vehicles, respectively host vehicle 2 and neighbouring vehicle 10, at a later time instant T _k+ i>T _k . The period of time [T _k ,T _k+ i] forms an observation time interval, wherein n measurements are being captured by the sensors on-board of vehicle 2 during the observation time interval.

The position information items representative of the measured positions of the travelling vehicles are depicted by dots in figure 2. For example, Q(T _k ) is the position of vehicle 2 at time instant T _k , and Q(t _kp ) is the position of vehicle 2 at instant t _k,p =T _k +(p-1 )x, p = 1 ,...,n. Similarly, R(T _k ) is the position of vehicle 10 at time instant T _k and R(t _k,p ) is the position of vehicle 10 at instant t _k,p =T _k +(p-1 )τ, p=1 ,.,.,η.

The reference paths, which are representative of the road lanes are schematically depicted as arrowed curves 12, 14 in figure 2.

The positions obtained from measurements show the deviation of each vehicle as compared to the respective travelled segments of reference paths 12, 14.

The precise reference paths are not known and not accessible to a vehicle.

In an embodiment of the invention, it is proposed to obtain an estimate of a reference path that a vehicle had already travelled during an observation time interval, and to compute a quantity of deviation of the real path of the vehicle from the estimated reference path.

Figure 3 represents a block diagram of various functional units of a system according to the invention.

A host vehicle 2 comprises one or several position and motion sensors 20 of the vehicle itself. In an embodiment, the sensors 20 comprise a speed sensor and a steering angle sensor, which are adapted to capture measurements at the approximately same time instants. Optionally, the host vehicle may comprise one or several sensors 22 to detect and track neighbouring vehicles, in particular adapted to capture the distance to and the speed of a neighbouring vehicle, as illustrated in figures 1 and 2.

The measurements provided by the sensors 20, 22 are stored in a memorizing unit 24 of a host vehicle 2.

The host vehicle 2 further comprises a reference values estimation unit 26, which obtains reference values for an observation time interval as explained in detail hereafter.

In a first embodiment, the reference values are computed from a subset of the memorized position and motion information, to estimate a reference path for a host vehicle 2 and a reference path for each neighbouring vehicle for an observation time interval. In this embodiment, an abnormal driving behaviour is meandering of the vehicle with respect to the estimated reference path.

In a second embodiment, a reference value is a maximum speed value allowed on a road or a speed value estimate of the traffic flow for the observation time interval. In this embodiment, a corresponding abnormal driving behaviour is excess speeding.

In a third embodiment, a reference value is a minimum range distance known as safe inter-vehicle headway between two vehicles travelling in the same lane and in the same direction, and maintaining a greater distance than said safe inter-vehicle headway ensures collision-free motion. In this embodiment, a corresponding abnormal driving behaviour is non respect of the safe inter-vehicle headway.

The reference values are stored in the memorizing unit 24.

Further, the host vehicle comprises a unit 28 for calculation of the sets of deviation values, where each set is representative of the corresponding driving behaviour and each deviation value corresponds to a measuring time instant of the observation time interval, as explained in more detail hereafter with respect to the various embodiments.

An assessment unit 30 provides the quantities of deviations from the respective reference values corresponding to the relevant driving behaviours during the observation time interval.

Indeed, if the sensors 22 are present and if one or several neighbouring vehicles have travelled in the neighbourhood of a host vehicle 2 during the observation time interval, then the assessment unit 30 computes the deviations quantities for the relevant driving behaviours of each neighbouring vehicle and, in relation to meandering of vehicles, a meandering matrix calculation unit 32 computes a meandering matrix for the host vehicle, which contains the deviation covariance of every two vehicles considered. Advantageously, the invention allows to apply awareness to neighbouring vehicles and to decrease the overall risk of accident on the road. The result of the assessment unit 30 and/or of the meandering matrix calculation unit 32, is transmitted to a decision making unit 34, which is adapted to check if a quantity of deviation for the relevant driving behaviour has exceeded a predetermined threshold during the observation time interval. The decision making unit 34 performs a statistical hypothesis testing according to a predetermined threshold probability known as a significance level, in order to decide whether the null hypothesis about a driving behaviour is rejected according to at least one significance level, for example 5% or 1 %, and therefore an abnormal driving behaviour is detected, or otherwise one cannot reject the null hypothesis given the observed data.

The decision making unit 34 is connected to a safety improvement unit 36.

The reference values estimation unit 26, the unit 28 for calculation of the sets of deviation values, the assessment unit 30, the meandering matrix calculation unit 32 and the decision making unit 34 are implemented by computing means, for example an onboard computer of a host vehicle 2.

In an embodiment, the safety improvement means 36 of a host vehicle 2 are adapted to raise an alarm, for example a visual alarm is displayed or an audio alarm is issued to alert the driver.

In an alternative embodiment, the safety improvement means 36 trigger a driver assistance system to take control of a host vehicle 2.

Figure 4 is a flowchart detailing the implementation of a first embodiment of the invention, wherein the reference values are estimates of the path travelled by the vehicle.

In a first step 40, a set of position information items Q(t _k,p ) defining the position of the host vehicle implementing the method are obtained and memorized, for a current observation time interval considered, [T _k , T _k+ i]. The position Q(t) of the vehicle at a given time instant t is a two-dimensional vector composed of the vehicle coordinates x(t), y(t) in a predetermined spatial referential. If the on-board sensors provide an information relative to the steering angle and speed at given time instants t _k , _p , the measured values are transformed into coordinates x(t _kp ), y(t _{k p} ) by calculation.

The position information items Q(t _kp ), with p varying from 1 to n, t _k,p varying from the starting time T _k = t _{k l} to the ending time T _k+1 = t _{k n} = t _k + (η - ϊ)τ of the observation time interval [T _k , T _k+1 ] are stored. The time instants t _k,p are called measuring time instants.

Similarly, the motion information is obtained at step 42. Preferably the motion information items are two-dimensional speed vectors Q'(t _{k p} ) of the host vehicle 2, at measuring time instants t _k,p .

Next, an estimation of the reference path travelled by the host vehicle for the observation time interval is computed in the curve fitting step 44. In an embodiment, the curve chosen to represent the path is a smooth spline P(t), also known as cubic spline, which is a continuous curve on the time interval [T _k , T _k+ i]. The derivative of the curve P(t), noted P'(t), is also continuous on the time interval [T _k , T _k+ i ].

The spline curve P(t) is fitted based on the boundary conditions, i.e. the measured position information values and motion information values at the starting and ending time instants, respectively T _k and T _k+ i , corresponding to the observation time interval. Therefore only a subset of the stored position information items and motion information items are used for the estimation of the reference curve.

In an embodiment, the smooth spline is defined by the following equation:

where t e [T _k , T _k+1 ] , k=0,1 This definition of a smooth spline is not limiting to the above formula and other definitions are possible.

The spline P(t) and the coefficients B . are two-dimensional vectors defining an estimate of the reference (x,y)-path travelled by the vehicle during the observation time interval [T _k , T _k+1 ]. The following and not limiting boundary conditions can be used to calculate the coefficients B _j :

P(T _k ) = Q(T _k ),

P T _k ) = Q T _k

P(T _k+l ) = Q(T _k+l ),

P (T _k+l ) = Q T _k+l )

Once the coefficients B _j are computed, the estimated reference path P(t _k,p ) or its respective position in the spatial referential considered at any time instant t _k,p of the observation time interval can be computed.

A set of deviation values ) - Q(t _{k >p} )| representative of the

deviation of the actual path travelled by the vehicle from the estimated reference path P(t), t e [T _k , T _k+l ] is next computed at step 46, for each p between 1 and n. The difference Q(t _{k p} ) = P(t _{k p} ) - Q(t _{k p} ) is a two-dimensional vector, and a corresponding deviation value |Δ< 0 _¾;Ρ )| is for example the L2 norm of said vector. Alternatively, another norm such as the L1 norm may be used. More generally, each deviation value is computed as a distance between a memorized position information item Q(t _k,p ) and the corresponding estimate reference position of the reference path P(t _k,p ). By construction, the respective deviation values at the boundaries of the observation time interval are equal to zero.

Alternatively, the reference path is represented by a smooth spline of a different order, or different boundary conditions are used.

In another alternative embodiment, the reference path is represented by a regression curve.

Next, a quantity of deviation of the vehicle from the estimated reference path is computed at step 48.

The sample mea computed as m _k (AQ) = et

of deviation values for the observation time interval [T _k ,T _k+1 ] is computed as:

The sample variance s _k ² (AQ) is an estimate of the vehicle meandering on the observation time interval [T _k , T _k+ i ]. In other words, the meandering is quantified by computing the sample variance of the vehicle deviation from a reference path estimate during a short time period.

Furthermore, another statistical measure of the amount of deviation for the set of deviation values |Δβ(ί _{¾ ρ} )| , such as the standard deviation, is used as an estimate of the vehicle meandering. Additionally, other statistical measures such as a sample skewness (known as a third moment) and a sample kurtosis (known as a fourth moment) can also be used to evaluate the vehicle meandering.

In an embodiment, the vehicle meandering is calculated using the sample variance of the set of deviation values, while the sample skewness and the sample kurtosis of the set of deviation values are supplementory.

The quantity of deviation given by the sample variance s (AQ) is then compared to a predetermined threshold θι corresponding to the meandering behaviour.

In an embodiment, the threshold θι depends on the vehicle speed during the observation interval, since the higher the speed, the lower the tolerated deviation. For example, the threshold θι is obtained from measurements as a heuristic value or is represented by a non-linear function.

If the computed quantity of deviation given by the sample variance s _k ² (AQ) is greater than θι , then an alarm is raised in step 52. Whether or not an alarm is raised, step 50 is followed by a step 54 of incrementing the value k by 1 , therefore triggering the processing of a subsequent observation time interval. Step 54 is followed by step 40 already described.

In practice, the obtaining and memorizing of position and motion information at regularly spaced measuring instants is preferably carried out as a background task, and memorized information items, corresponding to an observation time interval already processed may be erased from the memory. The observation time intervals processed are successive and their duration may vary.

In an alternative embodiment, the observation time intervals are partially overlapped or spaced in time.

The sample variances obtained for multiple observation intervals may be used in combination for comparing with the predetermined threshold θι . Analogously, the values of sample skewness obtained for multiple observation intervals can be used in combination. Analogously, the values of sample kurtosis obtained for multiple observation intervals can also be used in combination.

Advantageously, even though the estimate reference path computed is approximate, using short successive observation time intervals brings a good estimation of meandering. For example, if the sensors provide position and speed information every 50ms, an observation period of 5 seconds allows acquiring a hundred measurements. If the vehicle travels at an average speed of 90km/h, the observation time interval corresponds to 125 metres travelled by the vehicle. Therefore, the proposed method allows monitoring the vehicle meandering on short travelled distances.

Preferably, the observation time interval is chosen so as to contain a statistically significant number of measurements, for example n≥ 30 and preferably n≥ 100 .

Further, given that the observation time intervals considered are short, the method further allows using position and motion information regarding neighbouring vehicles, which can be monitored by sensors. For example, some available lidars provide a maximal range of 200 metres, therefore a neighbouring vehicle 10 can be monitored if it remains at a distance smaller than 200 metres from the host vehicle 2.

Let us denote Veh ₀ a host vehicle such as vehicle 2 of figures 1 and 2 and Veh, a neighbouring vehicle of host vehicle 2, such as vehicle 10 of figures 1 and 2.

As illustrated in figure 5, steps analogous to steps 40 to 48 of figure 4 are implemented to obtain an estimated reference path travelled by the neighbouring vehicle Veh, (step 56), a set of deviation values for the vehicle Veh, (step 58) and a sample variance of deviation s ² _k (i) (step 60) for each observation time interval [T _k , T _k+1 ]. Other measures such as the sample skewness or the sample kurtosis of the set of deviation values can also be computed.

Given that a set of deviation values during the observation time interval is obtained for each vehicle and the meandering calculation is performed on-board of the host vehicle Veh ₀ , a meandering matrix for the host vehicle Veh ₀ can be constructed. The meandering matrix contains the sample variance of deviation values for the host vehicle and its neighbouring vehicles on its main diagonal, and the non-diagonal values of said matrix are covariances s ² _k (i,j) (step 62) computed between the respective sets of deviation values associated to vehicles Veh, and Veh , travelling on the road.

Therefore, the meandering matrix obtained (step 64) contains the sample variance of deviation values on its main diagonal for the host vehicle and for each neighbouring vehicle if any, and the covariances represent the interdependence of motion between vehicle Veh, and vehicle Veh _j for each observation time interval. These covariances may serve, in particular, for safety evaluation of the vehicles travelling as a group on a road. If the covariance between the neighbouring vehicles Veh, and Veh is greater than a threshold θ ₂ (step 66), this indicates a degree of interdependence of said vehicles in relation to meandering, and an alarm is issued (step 68) to the driver of the vehicle that implements the proposed method. Optionally, the alarm may also be communicated to the drivers of vehicle Veh, and vehicle Veh _j in order to alert them.

According to a second embodiment, illustrated by the flowchart of figure 6, the reference value V _ref is a maximum speed allowed on the road where the vehicles are travelling or a speed of the traffic flow.

It is assumed that the motion information sensors on-board of a host vehicle provide instant speed information for said vehicle itself and for its neighbouring vehicles at each measuring time instant during the observation time interval. A first step 70 of the method consists in obtaining and memorizing a reference speed value V _ref for the observation time interval. Such a value may be obtained automatically from a driver assistance system or a vehicle-to-infrastructure communication system or, for example, be set manually by the driver according to the speed limit of the road. Next, a set of deviation values is computed at step 72 for each observation interval, that is to say each deviation value is calculated at time instant t _k,p =T _k +(p-1 )x, p=1 ,...,n according to the following formula: Advantageously, only the measured instant speed values which are greater than the reference speed value are actually taken into account.

Next, at step 74, a sample variance of speed deviation is computed for the set of deviation values {AV(t _{k p} ), p = l,...,n] for each observation time interval.

Next, step 78, analogous to step 50 previously described, is implemented, wherein the computed variance is compared to a threshold 6> ₃ which is a speed deviation variance threshold. Preferably, this threshold 6> ₃ depends on the reference speed value on said road. If the speed deviation variance computed is greater than the speed deviation variance threshold, an alarm is raised (step 80) analogously to step 52 previously described.

Optionally, the reference speed value V _ref is a speed value of the traffic flow where the host vehicle is moving, for example an average or mean speed of the traffic flow.

According to a third embodiment, the method aims at avoiding collision between a host vehicle and a neighbouring vehicle, both travelling in the same lane and in the same direction. In the third embodiment, the reference values are representative of a minimum safe range distance value between a host vehicle and a neighbouring vehicle.

In this third embodiment, the sensors of a host vehicle provide instant speed V(t _{k p} ) information of said vehicle and range distance D(t _{k p} ) between said vehicle and a neighbouring vehicle at measuring time instants t _k,p .

As shown schematically in figure 7, the host vehicle 2 is equipped with sensors which provide information on the neighbouring vehicle 10', travelling on the same lane 6 of road 4 as the host vehicle 2. The sensors provide a range distance D(T _k ) at time instant T _k between the host vehicle 2 and the neighbouring vehicle 10', travelling in front of host vehicle 2. To ensure safety, the distance D(t _k,p ) should be, at any time instant, above a minimum safe range distance between a host vehicle and a neighbouring vehicle travelling in the same lane and in the same direction ahead of the host vehicle, called a safe inter-vehicle headway. According to an embodiment, the safe inter-vehicle headway is given by the formula:

D _s (t _p ) = D ₀ + V(t _p )C _s ,

where D ₀ is a constant safety spacing between vehicles, C _s is a predetermined constant.

A set of deviation values from the instant minimum range distance is computed as follows: ,

Advantageously, a positive deviation value is computed only when the inter-vehicle distance D(t _k,p ) is smaller than a safe inter-vehicle headway D _s (t _{k p} ).

Once the set of deviation values is computed, as previously explained, a sample mean and a sample variance are computed, and the sample variance is used as a quantity of inter-vehicle headway deviation with respect to the safe inter-vehicle distance. The quantity of inter-vehicle headway deviation is compared to a corresponding heuristic threshold, and an alert can therefore be triggered if necessary.

Similarly to an example of figure 7, the neighbouring vehicle travels behind the host vehicle whose rear sensors provide an inter-vehicle headway between the host vehicle and its neighbouring vehicle.

According to an alternative embodiment, the sets of deviations AQ, AM and AD are computed for the host vehicle and for each of its neighbouring vehicles, so that a vector of variances is obtained for each vehicle.

In overall, the volatility of a travelling vehicle and a driving behaviour is associated with the variance of the vehicle deviation from the corresponding reference estimates. Therefore, the assessment of abnormal driving behaviour is achieved by observing and monitoring of the travelling vehicles and by evaluating a sample variance of the deviation between the actual motion parameters and their reference estimates during a short time period. The variance calculation of the relevant motion parameters is repeated iteratively to quantify the volatility of driving behaviour of a vehicle during subsequent time periods.