RU2486597C1 - Method of automatic classification of vehicles - Google Patents

Abstract

FIELD: transport.

SUBSTANCE: proposed method comprises seven main steps: (1) correction of radial distortion, (2) detection of entries and/or departures, (3) determination of vehicle height, (4) detection of vehicle wheel axles, (5) pre-calculation of vehicles axles' number, (6) filtration of pre-calculation results (statistical processing), and (7) final calculation of vehicles axles' number.

EFFECT: high-accuracy, absolutely independent multi-criterion classification of vehicles.

9 cl, 16 dwg

Description

FIELD OF TECHNOLOGY

The invention relates to means for the automatic classification of vehicles,

BACKGROUND

From the descriptions of published applications and patents US 5446291 A, US 5750069, US 3748443, WO 2005017853 A1, US 2002140924 A1, US 6195019 B1, US 5107250 A, WO 03052457 A2, WO 03094128 A2, FR 2903519 A1, US 2010231720 Al, EP 0807914 A1, US 5809161 A, DE 4300650 A1, EP 2026246 A1, US 6185314 B1, WO 2008088409 A2, WO 2011000677 A1, WO 9516252 A1 and WO 03096300 A1, known for their different functional capabilities and technical features of the system that implement individual approaches to automatic classification of vehicles, as well as components of these systems.

Moreover, the aforementioned systems, known from publications US 5446291 A, US 5750069 A, US 3748443 A, WO 2005017853 A1, US 2002140924 A1, US 6195019 B1, US 5107250 A, WO 03052457 A2, WO 03094128 A2, FR 2903519 A1, are based on the use of technically sophisticated equipment - optocouplers and rangefinders, tire pressure sensors, electrical sensors, optocouplers and a raster profile scanner, laser scanner, seismic sensors, temperature sensors - and besides, using these systems it is usually impossible to determine the number of axles and the height of vehicles funds.

The mentioned systems US 2010231720 A1, EP 0807914 A1, US 5809161 A, DE 4300650 A1, EP 2026246 A1, US 6185314 B1, WO 2008088409 A2, WO 2011000677 A1, WO 9516252 A1, WO 03096300 A1 do not allow to calculate the number of axles of the vehicle, or they do not determine its height, or are not applicable for automatic classification for other reasons.

The closest analogue to the invention, from the point of view of the approaches used, is a classifier known from the international publication WO 03096300 A1. It relies mainly on the ratio of the radii of tires and wheels. However, as was experimentally established, for typical vehicles used in the territory of the Russian Federation, there is no reliable correlation between the classes of vehicles and the radii of tires and wheels, which does not allow the use of this classifier for the automatic classification of vehicles.

SUMMARY OF THE INVENTION

The objective of the present invention is to automatically classify vehicles according to the criteria adopted in the Russian Federation, using standard cameras, without the need for complex and non-standard sensor equipment.

The technical result is a high-precision, fully autonomous (without human intervention) multicriteria classification of vehicles (in height and number of axles), or rather, a method that ensures the receipt of these results in its implementation.

In accordance with the present invention, in the proposed method for the automatic classification of vehicles (hereinafter, the method, unless the context requires otherwise), digital optical-electronic converters based on a video camera and digital data processing means are used, and a video stream is continuously received from a camera mounted to the side of the observation zone compensate the radial distortion of the video stream of frames to obtain a plurality of values of the elements P _xyt stream where x - horizontal coordinate, y - vertical coordinate, t - Vreme co-ordinate; calculate the exponential moving average (over time) of the elements of the video stream B _xyt ; calculate the correlation coefficient at the coordinates x, y D ₁ (t) between P _xyt and B _xyt and determine such times t _in when D ₁ (t) becomes greater than a predetermined threshold value L ₁ , after which such moments t _out when D ₁ (t) becomes less than L ₁ ; calculate integrated and / or time-averaged P _xyt for the time period T ₄ (the period the vehicle is in the camera’s field of view) between the moments t _in and t _out for the set of constants y, obtaining the set P _T4 (y); calculate the difference D ₂ (y, t) between P _xyt (y, t) and P _T4 (y) and find such y _top (t) (corresponding to the vehicle profile) at which the values of D ₂ (y, t) exceed previously predetermined threshold value L ₂ ; recognize elliptical contours on separate frames (subsets of P _{xyt of the} form P _t (x, y)) of the video stream corresponding to the period T ₄ , determine the coordinates of the centers of the ellipses (X _c (t), Y _c (t)), discard the values, the vertical coordinate Y _c (t) of which differs by an amount greater than the threshold value L ₃ from the averaged values, obtaining a set of filtered values of the coordinates of the centers of the ellipses (X _cf (t), Y _cf (t)); apply the Hough transform to X _cf (t), and the number of axles of the vehicle is determined by the number of maxima in the Hough space; the values of Y _cf (t) and the maximum value of y _top for the period T ₄ determine the height of the vehicle.

Thus, the proposed method allows, without human intervention and the use of sophisticated technology, to determine all the criteria necessary for classifying a vehicle to one or another class provided for by the classification adopted on the territory of the Russian Federation.

In one preferred embodiment, the position of the images of the wheels of the vehicles in the image is determined using the Viola-Jones method.

In another preferred embodiment, the additional filtering of the values of the vertical coordinate, taking into account the permissible size of the wheels of vehicles.

In a particular embodiment, said cameras are mounted at different vertical levels.

In yet another particular embodiment, at least two cameras are used whose field of view is at least partially overlapped.

In a preferred embodiment, at least two cameras having different sensitivity of the sensors are used.

In one preferred embodiment, at least two cameras are used, and they are oriented in the longitudinal plane so that the optical axes are angled towards each other and the side borders of the camera’s fields of view intersect near the borders of the detection zone without leaving beyond its bounds. The intersection of the cameras' field of view forms a combined detection zone with such parameters that the video stream received by individual cameras from objects closer and further from the detection zone can be filtered depending on whether the object is present in the combined detection zone or not.

In an alternative preferred embodiment, at least two cameras are used, while one of them is oriented at an angle to the horizontal so that the border of the field of view of this camera is located near the border of the detection zone, without going beyond it.

In one particular embodiment, a lighting device is used to increase the illumination of the detection zone with a lack of illumination.

The proposed method allows you to automatically determine the class of vehicles (vehicles) passing the control section of the road lane, based on measuring the height and number of wheel axles of the vehicle, using exclusively data obtained from video cameras.

DETAILED DESCRIPTION OF THE INVENTION

As follows from the foregoing, the proposed method consists of seven main stages: (1) correction of radial distortion; (2) detection of entries and / or exits; (3) determining the height of the vehicle; (4) vehicle axle detection; (5) a preliminary calculation of the number of axles of the vehicle; (6) filtering the results of the preliminary calculation (statistical processing); and (7) the final calculation of the number of axles of the vehicle. Auxiliary steps of the method are: collecting information, modulating-demodulating signals, storing data, transferring data between nodes of the system. The main stages will be described in more detail below, by the example of particular forms of embodiment, and the features of the auxiliary ones are well known to those of ordinary skill in the art and do not need explanations.

CORRECTION OF RADIAL DISTORTION

We will define the model of the camera with radial distortion by equations describing the distortions of the two-dimensional image obtained by it with respect to the two-dimensional image, which would be obtained under the same conditions by an ideal pinhole camera. Radial distortion model (see, for example, Zhang Z. A Flexible New Technique for Camera Calibration // IEEE Trans. Pattern Anal. Mach. Intell. Washington, DC, USA: IEEE Computer Society, 2000. Vol.22, No. 11. P. 1330-1334) is described by the following dependence of the coordinates of a point in a real image with a distortion (x _d , y _d ) on the coordinates on

perfect shot (x _u , .y _u ):

$$\begin{array}{c}{x}_d=x_u+k_{\mathrm{one}}\cdot \left(x_u-x_c\right)\cdot \left(x_u^2+y_u^2\right)+k_2\cdot \left(x_u-x_c\right)\cdot {\left(x_u^2+y_u^2\right)}^2\\ y_d=y_u+k_{\mathrm{one}}\cdot \left(y_u-y_c\right)\cdot \left(x_u^2+y_u^2\right)+k_2\cdot \left(y_u-y_c\right)\cdot {\left(x_u^2+y_u^2\right)}^2\end{array}\left(\mathrm{one}\right)$$

In the formula (1) (x _c , y _v ) are the coordinates of the center of the radial distortion, ak _i are the distortion parameters.

With known distortion parameters for each pixel of an ideal image, its coordinates on the recorded (distorted) image are known. By sequentially copying the corresponding pixels, we obtain a reconstructed ideal image, with the exception of areas in which the corresponding coordinates extend beyond the boundaries of the distorted image (Figure 1).

ENTRY / EXIT DETECTION

The task of detecting vehicle entrances and exits is to determine the presence of a vehicle in the camera’s field of view in the studied (nearest, hereinafter referred to as its first) lane.

The approach to detecting entry-exit used in the proposed method is based on the geometric model of the scene shown in Figure 2 (the letters indicate the components of the scene: a is the control line, at the intersection of which we define events; b is the region of the far lanes of the road; c is the control detection zone trailer, d - control zone for determining the entry-exit, e - the area of the first lane). An event (entry or exit) is considered to be the intersection of the vehicle with an imaginary straight line passing through the center of the frame. The frame is divided into 2 parts - the region of the first lane (e) and the region of the far lanes (b). To detect an event, only the region of the first lane is used, since in the region of the far lanes there can be vehicles traveling along any other (not of interest to us) lane.

An approach is used to determine the presence of vehicles using a background image, which is calculated by applying a recurrent exponential filter to the video stream:

$$\begin{array}{c}B\left(t_n\right)=\beta \left(t_n\right)B\left(t_n\right)+\left(\mathrm{one}-\beta \left(t_n\right)\right)F\left(t_{n-\mathrm{one}}\right)\\ \beta \left(t_n\right)={\left(\mathrm{one}-\alpha \right)}^{t_n-t_{n-\mathrm{one}}},\alpha <<\mathrm{one}\end{array}\left(2\right)$$

In the formula (2), B (t _n ) is the value of the background image pixel at time t _n , F (t _n ) is the value of the corresponding image pixel from the camera at time t _n ; α is estimated based on the length of the maximum allowable time period, as well as on the idea of the time scale of changes in the true background image.

The background image B (t, x, y) thus obtained is compared with the camera image F (t, x, y) by the Pearson correlation coefficient;

$$d\left(t\right)=\frac{〈BF〉-〈B〉〈F〉}{\sqrt{〈B^2〉〈F^2〉}},{〈B〉}^{def}=\frac{{\displaystyle \sum _{x=0y=0}^{x_{\max },y_{\max }}B\left(x,y\right)}}{\left(x_{\max }+\mathrm{one}\right)\left(y_{\max }+\mathrm{one}\right)}\left(3\right)$$

As a result, using formula (3), we obtain the time series d (t), which we further analyze for the presence of features (i.e., events of interest to us), after having performed smoothing to reduce the noise level. The presence of an object in this case corresponds to a low value of d (t).

When determining the entry-exit by the value of d (t), a threshold detector with hysteresis is used: entry is recorded at d (t) <h ₂ , and exit is recorded at d (t)> h ₁ , and h ₂ > h ₁ .

In order to avoid an undesirable break of the vehicle on the tractor and trailer, the algorithm for determining the entry-exit relies on an additional control zone (see Figure 2). Assuming movement from left to right, the presence of an object in this control zone means that after a while the object will appear in the main control zone (see Figure 2). Combining data from two control zones, we obtain entry and exit criteria that are more resistant to trailers: entry is registered when an object appears in the main control zone, and exit is recorded when there is no object in both control zones.

HEIGHT DETERMINATION

Vehicle height is defined as the distance from the road plane to the highest point of the vehicle belonging to the body or rigidly fixed to the part body.

The task of determining the height falls into 2 parts: determining the vertical boundaries of the vehicle in the image and calculating the actual height of the vehicle based on the vertical boundaries.

, которая затем сглаживается гауссовским фильтром для устранения влияния шума на изображении. Сглаженное гауссовским фильтром изображение берется по модулю, и полученное таким образом изображение используется для поиска вертикальных границ ТС. Для обеспечения высокой производительности данного фильтра гауссовский фильтр реализован методом Дерише (см. Deriche R. Recursively Implementing the Gaussian and its Derivatives // Proceedings of the 2nd International Conference on Image Processing. 1992. P.263-267).The method of measuring the height is based on comparing the current image from the camera with the image averaged over the travel time of the vehicle with preliminary filtering by a gradient filter. Before processing, a rectangular fragment is selected from the original video stream, having a width of the order of 1/5 of the total frame width. Since each vehicle necessarily passes through the entire frame, fragment selection does not affect the quality of the results, but reduces the amount of computation necessary in proportion to the ratio of the fragment area to the area of the full frame. The gradient filter works according to the following scheme: based on the original image F (x, y), the approximation of the vertical gradient is calculated $$\frac{\partial F\left(x,y\right)}{\partial y}\approx F\left(x,y\right)-F\left(x,y-\mathrm{one}\right)$$

, which is then smoothed by a Gaussian filter to eliminate the effect of noise on the image. The image smoothed by a Gaussian filter is taken modulo, and the image thus obtained is used to search for the vertical boundaries of the vehicle. To ensure the high performance of this filter, a Gaussian filter is implemented using the Deriche method (see Deriche R. Recursively Implementing the Gaussian and its Derivatives // Proceedings of the 2nd International Conference on Image Processing. 1992. P.263-267).

f(y, t). Полученный f(y) интегрируется во времени путем взятия максимума: $$f_r\left(y,t\right)=\underset{t\in \left[t_{въезд},t_{выезд}\right]}{\max }f(y,t).$$

As the field of the indicator of the presence of the object O (x, y, t), the difference module of the image averaged over the travel time (transformed by the gradient filter) and the (transformed) image at the current time is used. In this case, O (x, y, t) is updated on each frame, which allows us to further integrate the spatial signal f (y, t) in time. The spatial signal itself is calculated as a line maximum: $$f\left(y,t\right)=\underset{x}{\max }O\left(x,y,t\right)$$

f (y, t). The resulting f (y) is integrated over time by taking a maximum: $$f_r\left(y,t\right)=\underset{t\in \left[t_{\mathrm{at}besd},t_{\mathrm{at}sesd}\right]}{\max }f(y,t).$$

The vertical boundaries of the vehicle in the image are calculated by applying the threshold detector previously described in this section to f _r (y).

To obtain the height of the vehicle in meters along the vertical boundaries of the image, a geometric model of the scene is used, assuming the horizontal optical axis of the camera and the perpendicularity of the optical axis of the vehicle. Schematically, this model is depicted in Figure 3 (where 1 is the road, 2 is the camera, 3 is the projective plane, 4 is the vehicle).

Let h ₀ be the height of the optical axis of the camera above the ground (a known parameter), y ₁ and y ₂ be the vertical boundaries of the vehicle in the image, and h is the desired height of the vehicle. We also assume that the image recorded by the camera is equivalent to the central projection of the scene onto a plane perpendicular to the optical axis of the camera and located at a distance d0 from the camera. The following formula for h is obtained:

$$h=h_0\frac{y_2-y_{\mathrm{one}}}{y_2-y_0/2}\left(\mathrm{four}\right)$$

In the formula (4), y ₀ is the vertical coordinate of the horizon line in the image.

It should be noted that the determination of the lower boundary of the vehicle is based on the coordinates of the wheels found on the image by the Viola-Jones method (Viola P., Jones M. Robust Real-time Object Detection // International Journal of Computer Vision. 2002).

AXIS VEHICLE DETECTION

The task of determining the number of axles of the vehicle directly reduces to the task of searching (or recognizing) all kinds of wheels on each frame of the video sequence between the entry and exit of the vehicle with subsequent integration of the results. To increase reliability, several methods of searching for wheels are used.

In Fig. 4, Fig. 5 and Fig. 6, several examples of images of wheels of cargo and passenger vehicles are illustrated. The examples show the possibility of statistically (with automatic mode in the presence of a sufficiently large training sample) to train wheel detectors for trucks and cars that can operate in a fairly wide range. To construct statistical detectors, the Viola and Jones scheme is used (see the decree. Op.).

Alternatively, an approach based on filtering elliptical edges can be used to recognize vehicle axles.

Figure 7 presents the original image of the zone of interest in the search for the axes of the vehicle.

Step 1. Find the edges in the image, according to the approach_Canny (Fig.8).

Step 2. We will remove all the branches on the border map, after which we will filter the border segments by size (delete both too short and too long segments) (Fig. 9).

Step 3. We remove the points of the segments of the edges in the vicinity of which the curvature of the segment is above the threshold. As a curvature estimator, we will use the angle formed by a two-link polyline consisting of the current point and two points with a serial number in a segment that differs by a small constant. Repeatedly filter along the length of the segments (Figure 10).

Step 4. We delete obviously straight-line segments. To do this, we calculate the inertia ellipse of each segment and discard the segments with the eccentricity of the inertia ellipse above the threshold (Fig. 11).

Step 5. We approximate each segment of the second-order curve. To do this, we will solve a system of the form:

$$\{\begin{array}{c}a{x}_{\mathrm{one}}^2+b{x}_{\mathrm{one}}{y}_{\mathrm{one}}+c{y}_{\mathrm{one}}^2+d{x}_{\mathrm{one}}+e{y}_{\mathrm{one}}=\mathrm{one}\\ a{x}_2^2+b{x}_2{y}_2+c{y}_2^2+d{x}_2+e{y}_2=\mathrm{one}\\ ...\\ a{x}_n^2+b{x}_n{y}_n+c{y}_n^2+d{x}_n+e{y}_n=\mathrm{one}\end{array}\left(5\right)$$

We calculate the invariants of the resulting second-order curve and discard the segments whose approximation does not correspond to the ellipse. For the remaining segments, we calculate the center of the ellipse and the radius (as the arithmetic mean of the dimensions of the semiaxes). Drop segments with a radius too large and too small. The result of the algorithm is shown in Fig.12.

PRELIMINARY CALCULATION OF THE NUMBER OF AXES OF THE VEHICLE (STATISTICAL PROCESSING)

Wheel detection allows you to detect wheels in a static frame without taking into account the movement of the car. This means that all frames on which a passing vehicle is captured are processed independently, and each time the wheels are searched again. Along with the fact that the detection module almost always produces several responses on one wheel, it turns out that the result of the detection is a table of wheels, the number of entries in which is many times higher than the actual number of wheels (axles) of the vehicle, and to calculate the latter, it is necessary to analyze this table.

We call a set S = [x, y, t, r, w), where x is the abscissa of the center of the wheel on the frame, y is the ordinate of the center of the wheel on the frame, f is the ordinal number of the frame on which the operation occurred, r is the radius of the wheel in pixels, w is the confidence level of the detector (weight). The set of responses K = {S _i } will be called a set.

At the input to the integrator, set K is supplied, by which it is necessary to determine the number of wheels. We denote it by n and by K _{j we} mean the subset of operations from K corresponding to the jth wheel, where j takes values from 1 to n. By K ₀ we denote the triggering from K that did not fall into any of K _j , i.e. false positives.

The problem boiled down to the problem of determining the number of clusters formed by data in the response space. Note that operations from the set K ₀ do not form a single cluster, because false positives can occur on any frame in any position, and therefore it is useful to filter them out before clustering. Two phases of the algorithm are distinguished: data filtering and clustering.

FILTERING PRE-CALCULATION RESULTS

(STATISTICAL PROCESSING)

The data filtering method used in the integrator is median filtering by two parameters: the ordinate of the center of the wheel and the radius of the wheel.

The vehicle moves in a straight line, therefore the position of the wheel x = x (t) is a linear function. This means that if in the plane (x, t), which we will continue to call the xt-diagram, we mark the points that the wheel occupied during the passage, we get a straight line. However, because of this, the wheel detector has some spatial tolerance, i.e. gives triggering even with small shifts of the search window, on a real xt - diagram, the stripes will be blurry.

FINAL CALCULATION OF THE NUMBER OF AXES OF THE VEHICLE

(STATISTICAL PROCESSING)

Consider the xt diagram with weights, i.e. we assume that each operation contributes to the intensity J (t, x) a contribution equal to its weight:

$$J\left(t_0,x_0\right)={\displaystyle \sum _{\left\{S|t=t_0,x=x_0\right\}}w}\left(6\right)$$

We apply the Hough transform to the image J: H = H (J). The image H (a, b) (in some integer coordinates (a, b) will produce a set of brightly glowing areas. Fig. 13 shows a real xt diagram for a truck with five wheel axles and its Hough transform. In order to calculate the number of wheels , the number of local maxima on H (J) greater than a certain threshold is calculated.

KEY COMPONENTS OF THE SYSTEM USED FOR THE IMPLEMENTATION OF THE SUGGESTED METHOD

A system for implementing the proposed method includes at least means for converting electromagnetic radiation into a digital signal (sensor unit) and means for digital data processing (digital data processing unit).

The sensor unit of the system is a set of several specialized video cameras with partially overlapping fields of view. There are several options for the geometry of the relative positions of the cameras, using from 2 to 4 cameras.

Consider the various options for a 4-chamber sensor unit.

The “T-variant” of the configuration of the cameras implies the location of three cameras at the same level (~ 1 meter) and the fourth camera at a height of ~ 2.5 meters (above the middle of the bottom). The optical axis of the cameras are perpendicular to the plane of the side of the vehicle (Fig).

The “II-variant” of the configuration of the cameras implies the location of two cameras on the same level (~ 1 meter) and two cameras above the lower ones at a height of ~ 2.5 meters.

The optical axis of the cameras are perpendicular to the plane of the side of the vehicle (Fig.15).

The “W-Option” of the configuration of the cameras implies the location of three cameras at the same level (~ 1 meter) with a minimum offset directly above one another and the fourth camera at a height of ~ 2.5 meters (above them). The optical axis of the lower chambers are located at angles of 45, 90 and 135 degrees to the plane of the side of the vehicle (Fig.16).

The two-chamber configuration may be “I” - the configuration of the sensor unit, in which two cameras are located vertically, the optical axes are horizontal in the horizontal plane perpendicular to the road, and the vertical axis of the lower camera is horizontal or has a slight slope to optimize the detailed image of the vehicle, while the axis of the upper The camera has a significant upward slope for the most complete coverage of the height range.

As the main two-chamber configuration, the configuration of the sensor unit is used, in which the cameras are located on different sides of the observed band and are directed along one axis towards each other.

Depending on the optimal configuration of the location of the cameras, the type of mount is determined. Two main types of mounting are possible. In the first type, all cameras are placed on separate metal racks installed on the inter-island security island. In the second type of attachment, the lower chambers are attached directly to the barrier grid. In this case, the vehicle height determination chamber is mounted on the remote bracket to the support column A from the side opposite the cabinet.

Claims (9)

1. A method for the automatic classification of vehicles, which use digital optical-electronic converters based on a video camera and digital data processing means; continuously receive a video stream from a camera mounted on the side of the observation zone; compensate for the radial distortion of the frames of the video stream, obtaining a set of values of the elements of the video stream P _xyt , where x is the horizontal coordinate, y is the vertical coordinate, t is the time coordinate; calculate the exponential moving average (over time) of the elements of the video stream B _xyt ; calculate the correlation coefficient at the coordinates x, y D ₁ (t) between P _xyt and B _xyt and determine such moments of time t _in , when D ₁ (t) becomes greater than a predetermined threshold value L ₁ and then determine such moments of time t _out , when D ₁ (t) becomes less than L ₁ ; calculating integrated and / or time-averaged P _xyt for the time period T ₄ (the period the vehicle is in the camera’s field of view) between the moments t _in and t _out for the set of constants y, obtaining the set P _T4 (y); calculate the difference D ₂ (y, t) between P _xyt (y, t) and P _T4 (y) and find such y _top (t) (corresponding to the vehicle profile) at which the values of D ₂ (y, t) exceed previously predetermined threshold value L ₂ ; recognize elliptic contours on separate frames (subsets of P _{xyt of the} form P _t (x, y)) of the video stream corresponding to the period T ₄ , determine the coordinates of the centers of the ellipses (X _c (t), Y _c (t)), discard the values, the vertical coordinate Y _c (t) of which differs by an amount greater than the threshold value L ₃ from the averaged values, obtaining a set of filtered values of the coordinates of the centers of the ellipses (X _cf (t), Y _cf (t)); apply the Hough transform to X _cf (t), and the number of axles of the vehicle is determined by the number of maxima in the Hough space; the values of Y _cf (t) and the maximum value of y _top for the period T ₄ determine the height of the vehicle. 2. The method according to claim 1, in which the position of the images of the wheels of the vehicles in the image is determined using the Viola-Jones method. 3. The method according to claim 1, in which additional filtering of the values of the vertical coordinate is carried out taking into account the permissible size of the wheels of vehicles. 4. The method according to claim 1, in which the said cameras are installed at different levels vertically. 5. The method according to claim 1, in which at least two cameras are used, the fields of view of which are at least partially overlapped. 6. The method according to claim 5, in which at least two cameras are used having different sensitivity of the sensors. 7. The method according to claim 1, in which at least two cameras are used, while orienting them in the longitudinal plane so that the optical axes are directed at an angle towards each other, and the side borders of the cameras' field of view intersect near the zone boundaries detection, without going beyond it. 8. The method according to claim 1, in which at least two cameras are used, while one of them is oriented at an angle to the horizontal so that the border of the field of view of this camera is located near the border of the detection zone, without going beyond it. 9. The method according to claim 1, in which additionally use a lighting device to increase the illumination of the detection zone with a lack of illumination.

Images