KR100261065B1 - Vehicle classification method and apparatus using image analysis - Google Patents

Abstract

PURPOSE: An apparatus for discriminating car model by using image recognition and method thereof are provided to simplify implementation as well as maintenance of the system, and to adopt the system for drive-through toll gate system by adopting image recognition method. CONSTITUTION: The apparatus for discriminating car model by using image recognition includes two cameras, four frame memories(120-140), two calculator/controllers(150), two frame switches(270), and a transceiver. The cameras are implemented on each sides of estimated moving path to detect predetermined detectable area of the car and outputs digital image data including the detectable area with a predetermined time difference. The frame memories store two-frame digital image data output from each of the cameras in a frame unit in turn. The calculator/controllers control overall operation of the system, analyze digital image data stored in the frame memory currently hooked on to calculate predetermined screen widths of left and right side of each car, put the calculated screen widths of left and right screen sides to a conversion ratio of the screen width to the actual width so as to discriminate the car model by using the calculated car width. The frame switches couple two frame memories on a corresponding side to the camera and the calculator/controller sequentially under the control of the calculator/controller. The transceiver transmits and receives the data between the calculator/controllers on two sides.

Description

Vehicle type classification method and device by image detection

The present invention relates to an apparatus and a method for classifying various types of vehicles that run on a roadway, and in particular, a vehicle model based on an image detection system capable of distinguishing the types of vehicles passing through a toll gate by a non-contact detection method using an image camera. It relates to a classification method and apparatus.

In general, the toll collection method of toll roads such as expressways is divided into a closed type and an open type. The closed type calculates a fare based on the type of the vehicle and the driving distance, and the open type calculates a fare based only on the type of the vehicle. Therefore, it is necessary for unit offices such as toll gates to automatically distinguish the types of individual vehicles passing through the toll gate, and the car model data thus classified is used not only for toll collection but also as basic data for various counting. .

The vehicle classifier in such a conventional toll collection system is largely divided into contact and contactless. The general configuration of the contact system consists of a vehicle separator and a vehicle separator. The vehicle type separator usually consists of a loop detector or an optical detector. Classification of the vehicle is typically performed by combining the lubrication and the wheel width, the pressure sensor train embedded in the road is made by sensing the number pressed by the wheels of the traveling vehicle. However, in the case of the contact type, there is a problem that a loop type detector or an optical detector for vehicle separation is required in addition to the vehicle type separator. In addition, the pressure sensor train used as a vehicle classifier is expensive in itself, and the pressure sensor train and the loop type detector need to be buried on the road floor, which is expensive to install and maintain, and also easily contacts with the wheels of the vehicle. There is this.

In addition, in order to solve various problems that arise in the process of issuance of tolls and tolls, in recent years, a so-called involuntary toll collection system has been actively developed to allow to collect charges by wireless communication while the vehicle is in operation. However, the contact type vehicle classification device has a problem that it is difficult to be used in combination with such a toll collection system.

In view of these problems, a method of determining a vehicle type in a non-contact manner by image detection has been actively proposed. The simplest of the image detection methods is a method of determining the presence or absence of a vehicle by comparing a change rate of the current input image signal with only a road using a single camera. In this method, the camera is installed at a certain height on the road at an angle to the front of the road, and when it is determined that the vehicle exists, the number of passing vehicles is counted based on this, and the size of each counted vehicle is measured. The vehicle types are estimated and estimated, and finally, the fare for each vehicle is calculated according to the result of the classification.

However, according to this method, when the vehicle is very many, since the vehicles are observed in close proximity to each other in an image input to a single camera, the vehicle behind the vehicle is blocked by the preceding vehicle, which makes it difficult to accurately count the vehicle. As a result, the vehicle covered with the road cannot immediately adapt to the state of the road, which changes according to time, and there is a problem that a lot of errors occur in determining whether the vehicle is present.

A more complicated method is to apply a multi-target tracking technique to determine the existence of a vehicle while tracking a large number of moving vehicles. The method of classifying cars by the size is being studied. However, even in such a tracking method, there is a problem in that a vehicle belonging to the same type is often mistaken as a vehicle belonging to another type due to the difference in the size of the vehicle in the observed image due to the difference in the height of the vehicle. .

In addition, in order to increase the accuracy of classification of vehicles, a method of dividing the vehicle into two types, large and small, has been proposed as the passing vehicle obscures these markings with a black and white mark on the floor of the road. However, even in this manner, if the number of times of contact with the wheels of the vehicle increases with time, the markers on the floor become worn out and thus there is a problem in that the ability to classify the vehicle is deteriorated.

The present invention has been made to solve the above-described problems, adopting the image detection method, easy to install and manage, low cost, vehicle model by the image detection that can be easily employed in a toll collection system The purpose is to provide a classification method and apparatus.

Another object of the present invention is to overcome the problems of the conventional image detection method and by the image detection to accurately determine whether the vehicle passes in real time, whether the vehicle forward / backward determination and vehicle classification by a relatively simple configuration It is to provide a vehicle classification method and apparatus.

According to an aspect of the present invention, a method of classifying a vehicle by image detection includes: (a) an actual width between two reference points of a road and a distance between the actual width and the screen distance, respectively, for cameras on both left and right sides installed to capture an expected movement path of the vehicle; Obtaining and inputting a conversion ratio; (B) storing the vehicle type classification table according to the width of the vehicle in advance; (C) calculating a local correlation by comparing each input feature signal generated from the left and right full images with each one-dimensional background feature signal updated and stored until the last time; (D) determining boundary positions on the left and right sides of the vehicle based on the calculated local correlation; (E) calculating a screen distance between the determined boundary position of the vehicle and the corresponding side reference point, substituting the calculated screen distance into the conversion ratio to obtain an actual distance, and calculating the obtained actual distance between the reference points; Calculating the actual width of the vehicle by subtracting from the actual road width; And (f) classifying the vehicle type by finding the actual width of the vehicle obtained in the step (e) in the vehicle type classification table.

On the other hand, the vehicle type classification apparatus according to the image detection according to another feature of the present invention is installed symmetrically on both sides of the expected movement path to capture a predetermined detection area on the expected movement path of the vehicle digital image captured the detection region Two camera means for outputting data at predetermined time intervals; Two frame memories each cyclically storing two frames of digital image data continuously provided from the respective camera means, one frame each; It controls the overall operation of the device and analyzes the digital image data stored in the frame memory currently connected to obtain predetermined screen distances on the left and right sides of the vehicle, respectively. Calculation / control means for each of the two sides for calculating the width of the vehicle by substituting the conversion ratio between the distances and for classifying the vehicle type based on the calculated width of the vehicle; One frame switching means on both sides cyclically connecting two frame memories on the corresponding side to the corresponding camera means and the calculation / control means under the control of each calculation / control means; And means for relaying the transfer of data between the calculation / control means on both sides.

1 is a perspective view schematically showing the installation state in the vehicle type classification apparatus by video detection of the present invention;

2 is a view showing an installation state of a camera in a vehicle type classification apparatus according to the video detection of the present invention;

3 is an overall block diagram of an image processing unit in the vehicle type classification apparatus according to the video detection of the present invention;

4 is an overall flowchart of a method for classifying vehicles by image detection according to the present invention;

FIG. 5 is a flowchart illustrating a process of determining whether a vehicle is present in FIG. 4;

6a and 6b is a view showing a setting state of the detection window and the width measurement window for determining whether the vehicle forward / backward and pass in the present invention,

7 is a view for explaining a technique for converting a two-dimensional image to a one-dimensional image in the present invention,

8 is a view for explaining a technique for obtaining a local correlation between a current input image and a reference background image in the present invention;

9 is a view for explaining a median filtering technique in the present invention;

10 is a state transition diagram according to a process of confirming whether the vehicle forwards / reverses and passes in the present invention;

FIG. 11 is a diagram for explaining a condition in which a progress flag is set in FIG. 10; FIG.

12 is a view for explaining a condition in which a backward flag is set in FIG. 10;

13A and 13B are diagrams for describing a condition of increasing a value of a water passage coefficient of a vehicle by one in FIG. 10;

14A and 14B are diagrams for describing a condition of decreasing a passing water quantity coefficient value of a vehicle by 1 in FIG. 10;

FIG. 15 is a flowchart for explaining a vehicle model determining process of FIG. 4; FIG.

FIG. 16 is a flowchart for describing a vehicle boundary determination process of FIG. 15; FIG.

17 is a view for explaining a setting state of a shadow boundary determination window in FIG. 16;

FIG. 18 is a diagram for explaining a principle of obtaining a conversion ratio between an actual distance and a screen distance in the present invention.

*** Explanation of symbols for main parts of drawing ***

10: road, 12: left first reference point,

14: left second reference point, 16: right second reference point,

18: right first reference point, 20, 22: vertical support,

24: horizontal support, 30, 40: left and right cameras,

50: image processing board, 60: maintenance / maintenance computer,

70, 80: monitor,

90: vehicle, 110: A / D converter,

120-140: frame memory, 150: digital signal processor,

160: D / A converter, 170: frame switcher,

210: A / D converter, 220-240: frame memory,

250: digital signal processor, 260: D / A converter,

270: frame switcher, 300, 400: the left and right full image,

302, 402: width measurement window, 304, 404: camera adjustment window,

310: shadow boundary judgment window,

LD1, LD2: left vehicle detection window,

RD1, RD2: Right vehicle detection window

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the vehicle type classification apparatus and method according to the image detection of the present invention in detail, the device will be described first for convenience.

1 is a perspective view schematically showing an installation state in a vehicle type classification apparatus according to the image detection of the present invention, and FIG. 2 is a diagram showing an installation state of the camera in the vehicle type classification apparatus according to the image detection of the present invention. As shown in FIG. 1 and FIG. 2, in the vehicle type classification apparatus of the present invention, the left first reference point 12 and the right first reference point 18 are respectively located at left and right edges of the road 10 through which the vehicle 90 passes. ) And the left second reference point 14 and the right second reference point 16 are also displayed on the inner side of the road 10. Two measuring cameras 30 and 40 are also provided at a constant height of the edges on the left and right sides of the road 10. However, on the vertical plane with the road 10 so that the vehicle 90 passing continuously can be accurately separated and the road 10 can be captured as broadly as possible, that is, at least the first reference point on the side of the vehicle and the second reference point on the opposite side of the vehicle. It is preferable to install the cameras 30 and 40 so as to incline 5 ° -10 ° toward the inside of the road 10. This way both cameras and their respective reference points will be symmetrical with respect to the center of the road.

On the other hand, the image captured by the camera 30 is provided to the image processing board 40 at 30 frames per second, for example, in the image processing board 50, the image data provided from the left and right cameras 30 and 40 in this way. By performing the process in real time to determine whether the vehicle 90 passing through the detection area, whether to determine whether forward / backward, the number of passing number and class of vehicles and the like at high speed. Reference numerals 20 and 22 denote vertical supports for supporting the cameras 30 and 40, and 24 denote horizontal supports. Reference numeral 60 denotes a computer for maintenance and repair of the device, and 70 and 80 denote monitors for displaying images captured by the left and right cameras 30 and 40 and processed vehicle classification in real time, respectively. .

3 is an overall block diagram of an image processing unit in the vehicle type classification apparatus according to the image detection of the present invention. As shown in FIG. 3, the image processing unit according to the present invention converts an image signal of a frame unit provided from the camera 30 into continuous digital data of predetermined bits, for example, 8 bits. ), Three frame memories 120, 130, and 140 that cyclically store three frames of continuous digital image data, one frame each, and analyze the digital image data stored in the corresponding frame memory. Digital signal processor 150 (hereinafter simply referred to as DSP) for performing tasks such as determining whether the vehicle 90 passes through the detection area, determining whether it is forward or backward, and counting the number of passes, and so on. The D / A converter 160 converts the image data into an analog signal and outputs the same. Reference numeral 170 circulates each of the frame memories 120, 130, and 140 to the A / D converter 110, the DSP 150, and the D / A converter 160 under the control of the DSP 150. Frame switcher to connect to

In the above-described configuration, each of the frame memories 120, 130, and 140 may be implemented as static random access memory (SRAM) having a size of, for example, 512 * 512, at the present time t. Assuming that the A / D converter 110 is connected to the frame memory 120, the frame memory 130 is connected to the DSP 150 and the frame memory 140 is connected to the D / A converter 160. In this state, when the DSP 150 finishes processing the image data input to the frame memory 130 at the time t-1, the DSP 150 outputs a signal indicating the signal to the frame switcher 170. Accordingly, the frame switcher 170 cyclically switches the connection state of the frame memories 120, 130, and 140 as shown by the dotted lines. That is, the frame memory 130 that is currently connected to the DSP 150 is connected to the D / A converter 160 so that the stored data is converted into an analog signal, and the frame is currently connected to the A / D converter 110. The memory 120 is connected to the DSP 150 to process the data, and the frame memory 140 connected to the D / A converter 160 is connected to the A / D converter 110 so that the data is stored in the camera ( 30 is updated with digital data inputted from and converted by the A / D converter 110. As such, the image input from the left camera 30 passes through three stages of the circulation queue 120, 130, and 140, until it is reconstructed by a monitor (not shown). There will be a delay of two frames. This is very effective when it is necessary to process the real-time image and display the result while receiving the actual image through the external camera 30. In this structure, when the signal processing of the DSP 150 ends in a sufficiently short time, each cycle of the image input (time t), DSP processing (time t-1), and display (time t-2) is theoretically 30 [ frame / sec].

In the above, the signal processing system coming from the left camera 30 has been described. The signal processing system coming from the right camera 40 is similarly similar to the A / D converter 210, three frame memories 220, 230, and the like. 240, DSP 250, D / A converter 260, and frame switcher 270. Reference numeral 280 denotes a dual port RAM for communication between each of the DSPs 150 and 250. The two DSPs 150 transmit data transmitted and received through the dual port RAM 280. (250) Any one of the DSP as a master to process the width of the vehicle 90 and classify the vehicle based on it, and displays the processing result in real time through each monitor 70, 80 do.

In the above-described embodiment, an analog camera is described as an example of the cameras 30 and 40. When using a digital camera, a simple interface circuit will be required instead of the A / D converters 110 and 210. The D / A converters 160 and 260 are provided for visually monitoring the images captured by the cameras 30 and 40. In addition, a general purpose microcomputer may be used instead of the DSPs 150 and 250.

4 is an overall flowchart of a method for distinguishing a vehicle type by image detection according to an exemplary embodiment of the present invention, and the steps described below are performed under the subject of the DSP 150. As shown in FIG. 4, in step S100, the left and right images input from the cameras 30 and 40 are processed in real time and digitized for each frame. In this process, two detection windows (to be described later) for determining whether the vehicle is present and for determining whether the vehicle is forward or backward are set at specific positions of the entire left and right images input from the cameras 30 and 40. Next, in step S200, the existence of the vehicle is determined using the image input in the detection window thus set, and in step S300, whether the current vehicle is moving forward using the vehicle presence determination result in step S200. If not, determine if you are going backward. In step S400, the number of vehicles passing through is calculated using the previous vehicle presence determination result and the forward / reverse determination result. Finally, in step S500, the vehicle models are classified based on the calculated width of the vehicle by processing the images input from the left and right cameras 30 and 40.

For reference, look at the specifications according to the vehicle type currently classified by the domestic vehicle classification device is shown in Table 1 below.

Comparison of specifications by vehicle type (Unit: cm) Car type range full width Battlefield Height 0 species 140 323-336 140-192 Type 1 169-189 357-547 129-261 2 types 193-230 477-869 202-348 3 types 248-250 647-1200 287-371

As shown in Table 1 above, in the case of battlefield or height, overlapping occurs between types, but in case of full width, the difference is between 4cm and 21cm in size. It is possible to classify vehicles without departing. Therefore, the present invention adopts a full-body classification method.

FIG. 5 is a flowchart illustrating a process of determining whether a vehicle is present in FIG. 4, and FIGS. 6A and 6B are diagrams illustrating setting states of a detection window and a width measurement window for determining whether the vehicle is forward / reversed and passed. 7 is a view for explaining a technique for converting a two-dimensional image to a one-dimensional image in the present invention, Figure 8 is a view for explaining a technique for obtaining a local correlation between the current input image and the reference background image in the present invention 9 is a diagram for describing a median filtering technique in the present invention.

First, in step S100 of FIG. 5, as shown in FIG. 6A, a specific position of the entire image 300 input to the left camera 30 is preferably located at a position perpendicular to the expected movement path of the vehicle. 6 detection windows LD1 and LD2 are set and shown in FIG. 6B even in a specific position of the entire image 400 input to the right camera 40, preferably in a position perpendicular to the expected movement path of the vehicle. Two detection windows RD1 and RD2 are set. Herein, the interval between the two detection windows LD1 and LD2 and the interval between the two detection windows RD1 and RD2 are preferably set to be equal to or less than the minimum interval that can be maintained between successive incoming vehicles. In this embodiment, it is set to about 110 pixels.

The images input to each of the detection windows LD1, LD2, RD1, and RD2 set as described above are used as basic information for determining whether the vehicle is present and moving forward or backward. Such detection windows LD1, The reason for setting (LD2), (RD1), and (RD2) is to reduce the amount of data compared to processing the entire image 300, 400, and to determine whether the vehicle is present and moving forward or backward at high speed. Because it is advantageous. In addition, it is preferable to keep the vertical length N of each of the detection windows LD1, LD2, RD1, and RD2 as small as possible to reduce the calculation amount, while the horizontal length M is sufficient for the vehicle to enter. It is a good idea to set it to include all paths that exist. In this embodiment, the vertical length N is set to 10 pixels, and the horizontal length M is set to 290 pixels. Reference numerals 304 and 404 denote adjustment windows for adjusting the states of the left and right cameras 30 and 40, respectively.

Next, in steps S200 to 212, the zone / part of the vehicle is determined based on the images input to the detection windows LD1, LD2, RD1, and RD2. Only the window LD1 is described as an example.

In operation S200, first, a feature signal is generated by converting a two-dimensional image input to the detection window LD1 into a one-dimensional image. To this end, first, as shown in FIG. 7, the two-dimensional image input to the detection window LD1 is converted into a one-dimensional image by performing an integral projection in the vertical direction. This is because most of the noise can be eliminated due to the averaging effect by the additive projection, assuming that the main characteristic of the noise existing in the vertical direction is random. In addition, the amount of computation to be processed can be reduced. If the two-dimensional image is added and projected in the vertical direction to make the one-dimensional image, the amount of data to be processed can be reduced to 1 / N as compared to processing the two-dimensional image as it is.

Next, when the obtained one-dimensional addition projection signal is differentiated after processing with an appropriate low pass filter, a characteristic signal of an input image in which a complicated and minute noise signal component is removed is generated.

In operation S202, a local correlation coefficient is calculated by comparing the feature signal of the input image thus generated with the feature signal of the background image (hereinafter, also referred to as a reference signal) generated in advance. As described above, in the present invention, a local correlation coefficient is used as a premise for determining the existence of the vehicle. In this case, the vehicle can be distinguished by the texture information, thereby effectively distinguishing the vehicle from the shadow. The reason is that although the shadows have less brightness than the road, the texture of the road remains above a certain level.

The local correlation coefficient is calculated by dividing each of the input feature signal and the background feature signal by a local window L having a constant length and then calculating the correlation coefficient between the local window L signals corresponding to the same position. Is done in a way that is obtained. This yields the same number of local correlation coefficients as the total number of local windows. On the right side of Fig. 8, the resultant local correlation result column is shown.

In the present invention, as described above, a correlation coefficient between signals in the local window L is obtained. The reason for this is to use the local characteristic of the input feature signal for vehicle detection. In other words, the input feature signal may be mixed with the vehicle part and the background part together. If the entire feature signal is compared to determine the vehicle part and the background part from the input feature signal, this is the average of the sum of the vehicle part and the background part. It is difficult to make a clear discrimination because the signal is compared and the result reflects the unique characteristics of the vehicle and the background at the same time. However, as described above, if the independent comparison of each local window (L), even if the entire signal is mixed with the vehicle part and the background part, because the unique characteristics of the vehicle and the background only appears locally Clear judgment can be made.

When the local correlation coefficient is calculated in this way, in step S206, the local correlation coefficient is binarized using a threshold, and then the occupancy of the vehicle is first examined based on this. In this case, a high correlation coefficient is obtained for an input signal similar to a background signal due to the nature of the correlation coefficient, and a low correlation coefficient is obtained for a signal having a large difference from the background signal. Therefore, the similarity between the background signal and the input signal can be known by the correlation coefficient, and the localized window L can be divided into a vehicle part and a non-vehicle part by binarization by applying an appropriate threshold set in advance thereto. .

Next, in step S206, the presence of the vehicle is determined based on the spatial occupancy of the vehicle thus obtained. For example, as shown in Table 2 below, the occupancy of the point having a low correlation is divided into three stages. If it is 45 [%] or more, it is determined to be a vehicle. If it is 16 [%]-44 [%], it is uncertain. If it is less than 15 [%], it is determined as a pure background without a vehicle.

Judgment result Share[%] Vehicle presence 45-100 uncertainty 16-44 Pure background (vehicle absent) 15 or less

In this way, if the existence of the vehicle in each frame is determined primarily, in step S208, the determined result for each frame is collected by a predetermined number in relation to the passing speed of the vehicle and the like to determine the existence of the vehicle. If it is determined in step S208 that the vehicle exists, the process proceeds to step S210 and median filtering the result determined for each frame to determine the final vehicle presence according to the result. The reason for doing this is to reduce the errors that can occur in the result of the existence judgment determined by only one detection window input. In other words, if a specific part of a vehicle having characteristics similar to the background is input to the detection window, the frame will be judged as a background, but in most cases except for a very few parts, the parts of the vehicle that are input to the detection window have unique characteristics. Median filtering can eliminate this change in impulse detection results.

On the other hand, when the median filtering is applied, the median filtering window should be set properly. The reason is that if the window size is set too large, two consecutive vehicles may be misidentified as one. If is set too small, the effect of applying median filtering is reduced. In the embodiment of FIG. 9, the size of the median filtering window is set to three, and as a result of the primary determination on the left, it is determined that the vehicle exists in the black frame, but the corresponding frame on the right obtained by applying the median filtering It can be seen that the vehicle is determined to not exist. In addition, it can be seen that the vehicle does not exist in the frame marked with the X mark because the vehicle is not sure whether or not the vehicle exists.

Meanwhile, when calculating a correlation coefficient, a background feature signal is estimated and used as a comparison target with an input feature signal. In this case, the estimated background feature signal is used as a result of the determination of the existence of the vehicle up to the previous frame as shown in Equation 1 below. Can be selectively accumulated. To this end, if it is determined in step S208 that the vehicle does not exist, that is, if it is determined that the current input image is a pure background image, the process proceeds to step S212 and an appropriate weight is applied to the background feature signal estimated until the last time. To be stored after adaptive update. Using the cyclic filter can remove some noise signal components mixed in the background signal, and adaptively update the background feature signal as the background changes.

은 추정된 배경 특징 신호를 나타내고, 인자 t는 프레임 번호를 나타내며, 인자 m은 일차원 신호의 수평 방향의 위치를 나타낸다. 또한 f_t(m) 은 입력 영상에서 구한 특징 신호이고, α 는 가중치 상수이다. 이 α 값을 차량의 존부 판단 결과에 따라 적응적으로 조절함으로써 순수한 배경 영상일 경우에만 배경 특징 신호를 갱신한다.From here

Denotes the estimated background feature signal, factor t denotes the frame number, and factor m denotes the horizontal position of the one-dimensional signal. Also f _t (m) Is the feature signal obtained from the input image, α Is a weight constant. this α The background feature signal is updated only in the case of a pure background image by adaptively adjusting the value according to the vehicle presence determination result.

α For example, it can be set to 0.2 if it is a pure road, and 0 otherwise.

In this way, if the presence or absence of the vehicle in each of the detection windows (LD1), (LD2), (RD1), (RD2) is independently determined, in step S300 and step S400 whether the vehicle is forward / reversed and the number of passages Count.

FIG. 10 is a state transition diagram according to a process of checking whether the vehicle is forward / reversed and passed in the present invention. In the case where the vehicle 90 exists in D1 and D2, the state value 1 and the vehicle 90 do not exist. If not, the status value is 0. As shown in Fig. 10, in the present invention, there are four possible states according to the vehicle presence determination result of D1 and D2. That is, when the vehicle 90 does not exist in both D1 and D2 (D1D2 = 00), when the vehicle 90 exists only in any one of D1 and D2 (D1D2 = 01 or 10) and the vehicle in both D1 and D2. There are four types of cases where (90) exists (D1D2 = 11). However, since the D1 and D2 are positioned vertically in the vertical direction, if the vehicle 90 enters the normal speed, the result of the determination of the presence or absence is changed sequentially. Accordingly, by determining the forward / reverse in accordance with the transition of these four states, the prepared forward flag and the reverse flag are changed, and each flag value and the history of the input state are used to accurately determine the number of vehicles 90 that have passed. Can be calculated In FIG. 10, the dashed arrows indicate prohibited state transitions, that is, non-existent state transitions, and when such a situation occurs, the device is regarded as having an error and an alarm is taken.

FIG. 11 is a diagram illustrating a condition in which the forward flag is set in FIG. 10, and FIG. 12 is a diagram illustrating a condition in which the backward flag is set in FIG. 10. A thick arrow in each input image indicates a driving direction of the vehicle. Indicates.

First, as shown in FIG. 11, it is possible to determine whether the vehicle 90 moves forward or backward according to the change of the states of D1 and D2. Looking at each of these cases, the initial situation of the left side of which the states of D1 and D2 is 00 is shown first. If the state changes to 10 on the right side, it can be considered that the vehicle 90 starts to enter from the top of the screen. Once it is determined that the vehicle 90 has started entering, it is assumed that the vehicle is moving forward unless the states of D1 and D2 return to 00 again. In such a case, a forward flag indicating a forward situation is set. Such a forward flag value remains set even if it changes to another state unless the states of D1 and D2 return to 00 again. Then, when the state goes back to 00 again, i.e., the vehicle immediately reverses from the entry point, the forward flag is reset to return to the initial state.

Similarly, the reverse flag is set to indicate the reverse state. As shown in FIG. 12, when the states of D1 and D2 change from 00 to 01, which is the initial state, the reverse flag is set to indicate that the reverse state is reversed. In case of returning to the 00 state, the reverse flag is reset to return to the initial state. By looking at the forward flag and the reverse flag in this way, it is possible to determine whether the vehicle is moving forward or backward.

13A and 13B are diagrams for describing a condition of increasing a passing quantity coefficient value of a vehicle by 1 in FIG. 10, and FIGS. 14A and 14B are diagrams for describing a condition of decreasing a passing quantity coefficient value of a vehicle by 1 in FIG. 10. As shown in the drawing, a thick arrow in each input image indicates a moving direction of the vehicle.

First, when the vehicle 90 moves forward and completely exits D1 and D2, the coefficient value should be increased by one. To this end, in the state where the forward flag is set on the state transition diagram of FIG. 10, the states of D1 and D2 change from 01 to 00 as shown in FIG. 13A, or from 11 to 00 as shown in FIG. 13B. In this case, the passing vehicle coefficient is increased by one.

On the other hand, if the vehicle passes through again and exits, it is necessary to reduce the coefficient for the vehicle. As in the case of the forward, as shown in FIG. 14A, as shown in FIG. 14A, the state of D1 and D2 is reduced. Is changed from 10 to 00 or from 11 to 00 as shown in FIG. 14B, the passing vehicle coefficient is decreased by one. 13B and 14B are prepared by assuming that the vehicle is very fast, for example. As described above, according to the present invention, even in a situation in which D1 and D2 do not sequentially confirm the existence of the vehicle, they are accurate. You can count.

FIG. 15 is a flowchart for describing a vehicle model determining process of FIG. 4. As shown in FIG. 15, in step S510, first, it is determined whether the vehicle 90 is detected in the left and right line detection windows LD1 and RD1. In step S510, the vehicle 90 is detected. In the case, the process proceeds to step S520 to determine whether the vehicle 90 is first detected in the left and right rear detection windows LD2 and RD2. When the vehicle 90 is first detected in the post detection windows LD2 and RD2 in step S520, the process proceeds to step S530 to determine the boundary of the vehicle 90.

FIG. 16 is a flowchart for describing a vehicle boundary determination process in FIG. 15, and FIG. 17 is a diagram for describing a setting state of a shadow boundary determination window in FIG. 16. First, as shown in FIG. 16, the boundary determination of the vehicle is performed in step S531. The width measurement window 302 for finding the boundary of the vehicle 90 in the left and right full images 300 and 400 is input. 402 are set as shown in Figs. 6A and 6B, respectively. The reason for setting the width measurement window 302, 402 is to obtain more accurate information with a small amount of calculation, so it is preferable to reduce the size as much as possible, but the horizontal length (C) is the vehicle 90 and the road ( It is good to set it to the length which can capture at least one boundary of the road 10 at least in relation to the width | variety of 10).

Next, in step S532, the boundary candidates of the vehicle 90 are set by analyzing the images input to the width measurement windows 302 and 402. For example, the outermost boundary is used as the boundary candidate of the vehicle 90. Set it. Next, in step S533, the shadow boundary determination window 310 of a predetermined length A is set in the direction of the vehicle 90 from the set boundary candidate as shown in FIG. Addition projection in the direction to generate a one-dimensional input feature signal. Since the reason for the addition projection has already been described, the explanation is omitted here. Next, the generated one-dimensional image is passed through an appropriate low pass filter and then differentiated to generate a one-dimensional feature signal from which the influence of noise is removed.

Next, in step S534, the correlation is calculated by comparing the one-dimensional input feature signal generated in step S533 with a previously stored one-dimensional background feature signal. The one-dimensional background feature signal is input to the shadow boundary determination window 310 set at the same position as the shadow boundary determination window 310 with respect to the pure background when no shadow exists in the width measurement windows 302 and 402. The image is obtained by horizontal addition as shown in the right side of FIG. The reason for this is that when the background signal without shadow is compared with the background signal with shadow, the latter shadowed background signal is similar to the background signal without shadow. This is because it can determine whether it is a shadow area or a vehicle area.

In operation S535, the obtained correlation is compared with a predetermined reference value. If the correlation is greater than or equal to the reference value, the correlation is determined to be a shadow, and the process proceeds to operation S536 to determine the boundary candidate as the vehicle 90. ), That is, the shadow boundary determination window 310 is moved again in the direction of the vehicle 90 by a predetermined distance (A), and then the steps S533 or less are repeated. In this manner, when the correlation is less than the reference value in step S535, the process proceeds to step S537 to determine that the currently set boundary candidate is the actual boundary of the vehicle 90.

When the boundary of the vehicle 90 is determined in this way, the process proceeds to step S540 of FIG. 15 again to calculate the width of the vehicle 90. In order to calculate the width of the vehicle, the actual distance W _Ref and the screen distance between the left first reference point 12 and the right second reference point 16 of the road 10 are set at the initial value of the left camera 30. It is assumed that the conversion ratio between L) is calculated and input in advance in the form of the camera's line of sight angle? With respect to one pixel. Of course, even when the initial value of the right camera 40 is set, the conversion ratio between the actual distance W _Ref and the screen distance L between the right first reference point 18 and the left second reference point 14 of the road 10. Is calculated in advance in the form of the camera's line of sight (θ) with respect to one pixel, and in addition, the actual width (W _B ) between the left first reference point 12 and the right first reference point 18 of the roadway 10. You must also enter

FIG. 18 is a diagram for explaining a principle of obtaining a conversion ratio between an actual distance and a screen distance in the present invention. As shown in FIG. 18, the height H and the screen of the left camera 30 measured and known in advance of the actual distance W _Ref between the first reference point 12 on the left side and the second reference point 16 on the right side. Based on the screen distance L between the two reference points 12 and 16 of the image, the gaze angle θ of the camera 30 with respect to one pixel is calculated as in Equation 2 below.

In such a relationship, the camera for one pixel stored in advance by inputting the distance d _L between the left boundary position of the vehicle 90 obtained in the step S537 and the left first reference point 12 of the road 10 in advance. Using the line of sight angle θ of 30), it is obtained as in Equation 3 below.

d _L = Htan (kθ)

In this way, if the distance d _R between the right boundary position of the vehicle 90 and the right first reference point 18 of the road 10 is obtained, then these distances are _subtracted from the actual road width W _B. The width W _C of the vehicle 90 can be obtained. If this is expressed mathematically, Equation 4 below.

Wc = W _B -d _L -d _R

In this way, when the width W _C of the vehicle 90 is obtained, the process proceeds to step S550 to compare the vehicle type classification table stored in advance and classify the vehicle model for the current passing vehicle. On the other hand, when the vehicle 90 is not detected in the line detection windows LD1 and RD1 in step S510, the process proceeds to step S560 to update and store the current road image. If the vehicle 90 is not initially detected in the later detection windows LD2 and RD2 in step S520, the program returns to step S510.

The vehicle type classification method and apparatus by video detection according to the present invention are not limited to the above-described embodiments, and various modifications can be made within the range permitted by the technical idea of the present invention. For example, in the above-described embodiment, an example of classifying a vehicle type by the width of the vehicle may be obtained by the height of the vehicle and the overall length.

Meanwhile, the term D1 used in the claims refers to at least one of LD1 or RD1, and D2 refers to at least one of LD1 or RD1.

According to the method and apparatus for distinguishing a vehicle type according to the image detection according to the present invention as described above, it is easy to install and manage by using the image detection method, the cost is low, and can be easily adopted for the toll-free toll collection system. It has an effect. In addition, there is an effect that can overcome the problems of the conventional image detection method and accurately determine whether the vehicle passes, determine the forward / backward of the vehicle and classify the vehicle in real time by a relatively simple configuration.

Claims (21)

(A) obtaining and inputting a real width between two reference points of the road and a conversion ratio between the real width and the screen distance in advance with respect to cameras on both left and right sides installed to capture an expected moving path of the vehicle; (B) storing the vehicle type classification table according to the width of the vehicle in advance; (C) calculating a local correlation by comparing each input feature signal generated from the left and right full images with each one-dimensional background feature signal updated and stored until the last time; (D) determining boundary positions on the left and right sides of the vehicle based on the calculated local correlation; (E) calculating a screen distance between the determined boundary position of the vehicle and the corresponding side reference point, substituting the calculated screen distance into the conversion ratio to obtain an actual distance, and calculating the obtained actual distance between the reference points; Calculating the actual width of the vehicle by subtracting from the actual road width; And And (f) classifying the vehicle types by finding the actual width of the vehicle obtained in the step (e) in the vehicle classification table. The method of claim 1, wherein step (c) (C-1) setting respective width measuring windows at specific positions inside the entire left and right images which are currently input; And (C-2) adding and projecting the two-dimensional image input to the respective width measurement windows in the vertical direction to convert the two-dimensional image into a one-dimensional image to generate a one-dimensional input feature signal. Classification method. 3. The method of claim 2, wherein the step (C-2) further comprises: performing differentiation after removing the high frequency component of the signal converted into the one-dimensional image by the addition projection in the process of generating the one-dimensional input feature signal. Vehicle type classification method by image detection characterized by the above-mentioned. The method of claim 3, wherein step (d) (D-1) setting a width measurement window for measuring the width of the vehicle in the entire image on both the left and right sides; (D-2) setting a boundary candidate of the vehicle from an image input to the set width measurement window; (D-3) setting a shadow boundary determination window having a predetermined length from the boundary candidate toward the vehicle; (D-4) generating a one-dimensional input feature signal by adding and projecting an image input to the shadow boundary determination window in a horizontal direction; (D-5) The image input to the shadow boundary determination window set at the same position as the shadow boundary determination window with respect to the pure background when there is no shadow in the width measurement window is added to the horizontal direction to project one-dimensional background feature signal. Generating; (D-6) comparing the degree of correlation between the one-dimensional input feature signal and the one-dimensional background feature signal generated in steps (la-4) and (ra-5), respectively; And (D-7) As a result of the comparison, if the correlation is greater than or equal to a predetermined reference value, it is determined that the shadow is determined, and after resetting the shadow boundary determination window to the direction of the vehicle, steps (a-4) or less are repeated, and the predetermined reference value is determined. And if not, determining the current boundary candidate as the boundary of the actual vehicle. 5. The method of claim 4, wherein the step (D-4) further comprises the step of differentiating the high frequency component of the signal converted into the one-dimensional image by the addition projection in the process of generating the one-dimensional input feature signal. Vehicle type classification method by image detection characterized by the above-mentioned. The method of claim 5, (The-1) setting the line detection window D1 and the post detection window D2 at two points in the vertical direction inside the entire image on both the left and right sides; (The-2) determining whether a vehicle exists in each of the set detection windows D1 and D2; (The-3) constructing a state transition diagram according to a result of the presence and absence of the vehicle determination performed in each of the detection windows D1 and D2; And (-4) further comprising determining a number of forward / reverse and passing numbers of the vehicle by analyzing the state change in the configured state transition diagram, The step (d) is carried out after the vehicle is detected at least in each of the line detection windows D1 and after the vehicle is first detected in the later detection windows D2. Classification method. 7. The method of claim 6, wherein the input image is obtained by photographing the image from above and below the expected moving path of the vehicle. 8. The method of claim 7, wherein the distance between the line detection window (D1) and the rear detection window (D2) is set to have a distance less than the minimum adjacent distance of the vehicle. 9. The state transition diagram of claim 8, wherein When the presence of the vehicle is not detected in both the line detection window D1 and the rear detection window D2 (D1D2 = 00); When the presence of a vehicle is detected only in the line detection window D1 (D1D2 = 10); When the presence of the vehicle is detected only in the post detection window D2 (D1D2 = 01); And The vehicle type classification method according to the image detection, characterized in that it is configured based on four states (D1D2 = 11) when the presence of the vehicle is detected in both the line detection window (D1) and the post detection window (D2). 10. The method of claim 9, wherein the forward / backward judgment in the step (the-4) is performed. (The-41) pre-setting the forward flag and the reverse flag; (The-42) setting the forward flag when the state of the D1D2 changes from 00 to 10 and setting the reverse flag when the state of the D1D2 changes from 00 to 01; And (The-43) A vehicle type classification method according to image detection, characterized in that the step of determining the forward / backward based on the state values of the forward flag and the backward flag. 11. The method of claim 10, wherein the set state is maintained as it is unless the state of the D1D2 returns to 00 in the state where the forward flag is set, And maintaining the set state as it is unless the state of the D1D2 returns to 00 in the state where the reverse flag is set. 12. The method according to claim 11, wherein the coefficient of passing quantity in the step (the-4) is When the state of the D1D2 changes from 01 to 00 or from 11 to 00 while the forward flag is set, increases by one; And a method of reducing by one when the state of the D1D2 changes from 10 to 00 or from 11 to 00 while the backward flag is set. The method according to claim 12, wherein the state of the D1D2 is changed from 00 to 11 while the forward flag or the backward flag is not set, When the state of the D1D2 is changed from 01 to 10 or from 10 to 01, it is determined that an error has occurred and an alarm is issued. The method according to any one of claims 6 to 13, wherein in the step (the-2), the determination of the presence or absence of the vehicle in each of the detection windows D1 and D2 is performed. (The-21) generating a one-dimensional input feature signal by adding and projecting the two-dimensional image input to each of the detection windows D1 and D2 into a vertical direction by adding and projecting the two-dimensional image in a vertical direction; (The-22) calculating a correlation coefficient for each local window by comparing the one-dimensional input feature signal with a previously stored one-dimensional background feature signal for each local window; (The-23) applying the calculated correlation coefficient for each local window to a predetermined threshold and binarizing and then investigating a share of the vehicle according to the binarized result; And (24) The method of classifying vehicles according to image detection comprising the step of first determining whether a vehicle exists by substituting the surveyed occupancy rate into a previously stored table. 15. The method of claim 14, wherein the step (the-21) further comprises the step of differentiating the high frequency component of the signal converted into the one-dimensional image by the addition projection in the process of generating the one-dimensional input feature signal. Vehicle type classification method by image detection characterized by the above-mentioned. The method of claim 15, wherein the determination of steps (the-21) to the step (the-24) is performed on a frame basis. (The-25) if it is determined in step (the-24) that the vehicle exists, determining whether the vehicle is finally present by median filtering whether the vehicle is determined in each frame unit in time. And (The-26) If it is determined in step (the-24) that the vehicle does not exist, the background feature signal is updated by applying a predetermined weight to the one-dimensional input feature signal generated in the step (the-21). Vehicle type classification method according to the image detection characterized in that it further comprises the step of. Two camera means provided symmetrically on both sides of the expected moving path so as to capture a predetermined detection area on the expected moving path of the vehicle and outputting digital image data capturing the detection area at predetermined time intervals; Two frame memories each cyclically storing two frames of digital image data continuously provided from the respective camera means, one frame each; It controls the overall operation of the device and analyzes the digital image data stored in the frame memory currently connected to obtain predetermined screen distances on the left and right sides of the vehicle, respectively. Calculation / control means for each of the two sides for calculating the width of the vehicle by substituting the conversion ratio between the distances and for classifying the vehicle type based on the calculated width of the vehicle; One frame switching means on both sides cyclically connecting two frame memories on the corresponding side to the corresponding camera means and the calculation / control means under the control of each calculation / control means; And And a means for relaying the transfer of data between the calculation / control means on both sides. 18. The image detection according to claim 17, wherein each calculation / control means comprises a digital signal processor dedicated to calculation, and wherein any one calculation / control means becomes a master to distinguish a width and a vehicle type of a vehicle. Vehicle classification device by. 19. The apparatus of claim 18, wherein the relay means comprises a dual port RAM. 20. The apparatus of claim 19, wherein the apparatus comprises: one monitoring means on each side for monitoring the respective processing results of the respective digital signal processors; In addition, one frame memory is provided on each side to store three frames of digital image data continuously provided, one frame each. And each frame switcher cyclically connects each of the frame memories to the camera means, the digital signal processor and the monitoring means under the control of the digital signal processor. 21. The vehicle type discrimination apparatus according to any one of claims 17 to 20, wherein the camera means is provided on a vertical surface of an unexpected path of the vehicle.

Images