Pure vision-based automatic door anti-collision system for vehicle

Abstract

The application relates to the technical field of automatic vehicle door control, in particular to an automatic door anti-collision system based on pure vision, which comprises at least one camera arranged at a vehicle door frame and used for collecting environment images of the rear and side of the vehicle when the automatic vehicle door is automatically opened by a preset distance; a processor connected with the camera and a door control module, used for receiving the environment images collected by the camera, detecting and identifying the moving objects at the rear and side of the vehicle according to the environment images, obtaining a control signal of the automatic vehicle door, and sending the control signal to the door control module; and the door control module is used for controlling the automatic vehicle door to continue to open or stop according to the received control signal, so that the accuracy of the visual detection result in the safety control of the automatic vehicle door is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic door control, and particularly relates to an automatic door anti-collision system based on pure vision. BACKGROUND

[0002] In recent years, with the continuous development of automobile intelligence, automatic door systems have gradually become an important configuration of medium and high-end vehicles. Such systems can achieve automatic opening and closing of vehicle doors through sensing or remote control. At present, in order to avoid collision with external obstacles during opening or closing, an automatic door anti-collision system for vehicles appears, which mainly relies on radar sensors or ultrasonic sensors to detect and identify whether there are obstacles outside the vehicle door or in front of the vehicle. When no obstacle is detected, the vehicle door is automatically opened, otherwise, the vehicle door is automatically closed. However, due to the limited detection angle of radar and ultrasonic waves, there are limitations in identifying stationary objects and dynamic objects (such as pedestrians or cyclists) in the complex environment near the vehicle door. At the same time, the cost of this system is relatively high, the installation is complex, and additional hardware integration is required, which limits its scale application. In addition, radar sensors may have false positives or false negatives when dealing with close-range or highly reflective objects, which can easily lead to inaccurate control of the vehicle door.

[0003] It is known that pure vision technology mainly relies on cameras to capture image information of the surrounding environment, and processes and analyzes the images through computer vision algorithms to understand the surrounding environment. This technology route has a high cost advantage because the price of the camera is relatively low. In addition, pure vision technology has certain advantages when dealing with complex scenes because it can extract more feature information from images.

[0004] Therefore, in view of the above technical problems, it is necessary to provide an automatic door anti-collision system for vehicles based on pure vision, which can reduce the cost and improve the control accuracy of the automatic door of the vehicle. SUMMARY

[0005] Therefore, in view of the above technical problems, it is necessary to provide an automatic door anti-collision system for vehicles based on pure vision, which can reduce the cost and improve the control accuracy of the automatic door of the vehicle.

[0006] An automatic door anti-collision system for vehicles based on pure vision is provided in the embodiment of the present application, which comprises:

[0007] At least one camera is arranged on the door frame of the vehicle, which is used to collect the environmental images of the rear and side of the vehicle when the automatic door of the vehicle is automatically opened by a predetermined distance.

[0008] A processor, connected with the camera and the door control module, is configured to receive the environment image collected by the camera, detect and identify the moving object behind and on the side of the vehicle according to the environment image, obtain a control signal of the automatic door of the vehicle, and send the control signal to the door control module.

[0009] The door control module is configured to control the automatic door of the vehicle to continue opening or stop according to the received control signal.

[0010] Preferably, the detecting and identifying the moving object behind and on the side of the vehicle according to the environment image to obtain the control signal of the automatic door of the vehicle comprises:

[0011] At least two frames of environment images collected by any camera are obtained, and each frame of the environment image is subjected to grayscale processing to obtain a corresponding grayscale image.

[0012] For any grayscale image, a grayscale histogram of the grayscale image is constructed, and a target connected domain in the grayscale image is obtained according to the grayscale histogram and the RGB value of each pixel point in the grayscale image, wherein the target connected domain refers to a region including the moving object of the vehicle and the pedestrian.

[0013] The target connected domain in each grayscale image is obtained, the target connected domains in all grayscale images are divided according to the overlapping area between the target connected domains in adjacent grayscale images, a plurality of target connected domain sets are obtained, and for any target connected domain set, the object moving risk degree corresponding to the target connected domain set is obtained according to the area and grayscale distribution of the target connected domains in adjacent frames in the target connected domain set.

[0014] The control signal of the automatic door of the vehicle is obtained according to the object moving risk degree corresponding to each target connected domain set.

[0015] Preferably, the obtaining the target connected domain in the grayscale image according to the grayscale histogram and the RGB value of each pixel point in the grayscale image comprises:

[0016] The pixel point quantity corresponding to the longitudinal axis in the grayscale histogram is connected to obtain a quantity change curve, a peak point in the quantity change curve is obtained, and the suspected target pixel point in the grayscale image is obtained according to the local distribution feature of each peak point.

[0017] The color difference feature value of the corresponding pixel point is obtained according to the RGB value of each pixel point in the grayscale image, and the target connected domain in the grayscale image is obtained according to the grayscale difference and color difference feature value of each pixel point in the local neighborhood of each suspected target pixel point.

[0018] Preferably, the local distribution characteristics of each of the peak points are used to obtain the suspected target pixel points in the any gray image, including:

[0019] For any peak point, the maximum pixel point number and the minimum pixel point number are obtained on the number change curve, the difference between the maximum pixel point number and the minimum pixel point number is calculated, which is recorded as a number range, the difference between the pixel point number corresponding to the any peak point and the minimum pixel point number is calculated, which is recorded as a first difference, and the negative of the ratio between the first difference and the number range is used as the independent variable of a preset exponential function to obtain the pixel distribution proportion corresponding to the any peak point.

[0020] The left valley point and the right valley point of the any peak point are obtained in the number change curve, the curve between the left valley point and the right valley point is intercepted as a local sub-curve of the any peak point, the slope between each coordinate point on the local sub-curve and its adjacent coordinate point is calculated respectively to obtain a slope sequence, the absolute value of the difference between each two adjacent slopes in the slope sequence is calculated to obtain an average difference absolute value, and the average difference absolute value is used as the independent variable of a preset exponential function to obtain the local fluctuation characteristic value corresponding to the any peak point.

[0021] The sum of the pixel distribution proportion and the local fluctuation characteristic value corresponding to the any peak point is calculated to obtain the probability value that the any peak point meets the moving object characteristics, and if the probability value that the any peak point meets the moving object characteristics is greater than or equal to a preset probability threshold, the gray value corresponding to the any peak point is determined as a suspected target gray value.

[0022] All suspected target gray values in the any gray image are obtained, and the pixel points corresponding to all suspected target gray values in the any gray image are used as suspected target pixel points.

[0023] Preferably, the color difference characteristic value of the corresponding pixel point is obtained according to the RGB value of each pixel point in the any gray image, including:

[0024] For any pixel point in the any gray image, the absolute value of the difference between the R channel value and the G channel value, the absolute value of the difference between the G channel value and the B channel value, and the absolute value of the difference between the R channel value and the B channel value are calculated according to the R channel value, the G channel value and the B channel value of the any pixel point, respectively, the cumulative value of all the absolute values of the differences is obtained, and the cumulative value is normalized to obtain the color difference characteristic value of the any pixel point.

[0025] Preferably, the target connected domain in any of the gray scale images is obtained according to the gray scale difference and color difference characteristic value of each of the suspected target pixel points and each pixel point in the local neighborhood thereof, and comprises:

[0026] For any suspected target pixel point in any of the gray scale images, the feature similarity degree between the any suspected target pixel point and each pixel point in the preset neighborhood range thereof is calculated according to the gray scale difference and color difference characteristic value difference, and if any feature similarity degree is greater than or equal to a preset feature similarity degree threshold, the pixel point corresponding to the any feature similarity degree is taken as a new suspected target pixel point.

[0027] The step of the any suspected target pixel point is repeated with the new suspected target pixel point as the any suspected target pixel point until all feature similarity degrees are less than the preset feature similarity degree threshold, and all new suspected target pixel points are obtained.

[0028] All new suspected target pixel points corresponding to each suspected target pixel point in the any gray scale image are obtained, and all suspected target pixel points and all new suspected target pixel points in the any gray scale image are taken to form at least one target connected domain.

[0029] Preferably, the feature similarity degree between the any suspected target pixel point and each pixel point in the preset neighborhood range thereof is calculated according to the gray scale difference and color difference characteristic value difference, and comprises:

[0030] Any pixel point in the preset neighborhood range of the any suspected target pixel point is taken as a neighborhood pixel point, and the absolute value of the difference between the gray scale value difference and the color difference characteristic value of the any suspected target pixel point and the neighborhood pixel point is calculated.

[0031] The addition value between the reciprocal of the absolute value of the gray scale value difference and the reciprocal of the absolute value of the difference of the color difference characteristic value is calculated to obtain the feature similarity degree between the any suspected target pixel point and the neighborhood pixel point.

[0032] Preferably, the target connected domains in all gray scale images are divided to obtain a plurality of target connected domain sets according to the overlapping area between the target connected domains in adjacent gray scale images, and comprises:

[0033] For the mth target connected domain in the last gray scale image, the last gray scale image is obtained, and the target connected domain with the largest overlapping area with the mth target connected domain in the last gray scale image is obtained, which is recorded as the same target connected domain of the mth target connected domain.

[0034] The same target connected domain of the mth target connected domain is taken as the mth target connected domain in the last gray-scale image, and the acquisition method of the same target connected domain of the mth target connected domain is repeated until the same target connected domain of the mth target connected domain in the first gray-scale image is obtained, and the mth target connected domain and all the same target connected domains of the mth target connected domain are combined to form an mth target connected domain set.

[0035] Preferably, the object moving risk degree corresponding to each of the target connected domain sets is obtained according to the area and the gray-scale distribution of the target connected domains of adjacent frames in the target connected domain set, and the method comprises the following steps:

[0036] The variance of the gray-scale values of all the pixel points in each target connected domain in the target connected domain set is obtained respectively, and is recorded as the gray-scale value variance of the corresponding target connected domain; the absolute value of the area difference and the absolute value of the difference of the gray-scale value variances between any two adjacent target connected domains in the target connected domain set are obtained respectively, and the addition value between the absolute value of the area difference and the absolute value of the difference of the gray-scale value variances is calculated, which is recorded as the inter-frame difference feature value between the two adjacent target connected domains.

[0037] The inter-frame difference feature values between every two adjacent target connected domains in the target connected domain set are accumulated to obtain the object moving risk degree corresponding to the target connected domain set.

[0038] Preferably, the control signal of the automatic door of the automobile is obtained according to the object moving risk degree corresponding to each of the target connected domain sets, and the method comprises the following steps:

[0039] A preset object moving risk degree threshold value is obtained, if the object moving risk degree corresponding to any target connected domain set is greater than or equal to the object moving risk degree threshold value, it is determined that the control signal of the automatic door of the automobile is stop; if the object moving risk degrees corresponding to all the target connected domain sets are all less than the object moving risk degree threshold value, it is determined that the control signal of the automatic door of the automobile is start.

[0040] Compared with the prior art, the embodiment of the application has the following beneficial effects:

[0041] When the automatic door of the automobile is automatically opened by a preset distance, the environment image collected by the camera is analyzed frame by frame, the moving objects behind and on the side of the automobile are detected and recognized through the environment image, the control signal of the automatic door of the automobile is obtained, and the control signal is sent to the door control module to control the automatic door of the automobile to continue to open or stop, so that the automatic door of the automobile is opened when the risk degree is high, and the accuracy of the safety control of the automatic door of the automobile is improved. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0043] Figure 1 is a structural block diagram of an automatic door anti-collision system of a vehicle based on pure vision provided by the first embodiment of the present application.

[0044] Figure 2 is a schematic diagram of automatic door opening on one side of a vehicle provided by the present application. DETAILED DESCRIPTION

[0045] The embodiments of the present disclosure will be described in detail below, and examples of the embodiments are shown in the drawings. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present disclosure, and cannot be understood as a limitation of the present disclosure.

[0046] It should be noted that the terms "first", "second" and the like in the specification of the present disclosure and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Rather, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0047] In order to illustrate the technical solutions of the present application, the following will be described by specific embodiments.

[0048] Referring to Figure 1 is a structural block diagram of an automatic door anti-collision system of a vehicle based on pure vision provided by the first embodiment of the present application, as Figure 1 shown, the system can include:

[0049] At least one camera 1 is arranged on the door frame of the vehicle, which is used to collect the environment images of the rear and side of the vehicle when the automatic door of the vehicle is automatically opened by a predetermined distance.

[0050] The camera is a camera mounted on the door frame of the vehicle, which includes a left camera and a right camera, and is used to monitor the moving objects on the left and right sides of the vehicle to assist safe driving. Considering that the images collected in heavy fog weather are difficult to identify moving objects, there will be false detection and false judgment, which will cause the automatic door to open in an unsafe condition, therefore, in the present embodiment, the Figure 2Taking the scenario that the automatic door on the side of the vehicle is opened as an example, that is, taking the environment image collected by the camera 1 on the side of the vehicle as an example, when the automatic door of the vehicle is automatically opened by a preset distance, a plurality of continuous environment images collected by the camera 1 on the side of the vehicle are acquired, and the environment images are transmitted to the processor, and image processing is performed by using the processor to observe whether there are moving objects such as vehicles, pedestrians and obstacles on both sides of the vehicle, so as to avoid dangerous accidents caused by the automatic opening of the automatic door when there are moving objects.

[0051] It should be noted that the preset distance is set to 3 cm or 5 cm, which is not limited here and can be set according to requirements.

[0052] The processor is connected with the camera and the door control module, and is used for receiving the environment image collected by the camera, detecting and identifying the moving objects behind and on the side of the vehicle according to the environment image, obtaining a control signal of the automatic door of the vehicle, and sending the control signal to the door control module.

[0053] After the processor receives the environment image, each frame of the environment image is subjected to grayscale processing to obtain a plurality of continuous grayscale images, so as to improve the detection accuracy of the moving objects in heavy fog weather through the image data analysis below. The grayscale processing belongs to the prior art and will not be described here.

[0054] The clothes worn by the pedestrian and the objects carried or ridden by the pedestrian usually have certain differences from the surrounding environment, especially under the influence of heavy fog weather, which will form a greater difference with the surrounding gray environment. In the gray histogram, the pixel points constituting the moving objects will gather together to form a curve peak with a slight difference and lower than other pixel values. Therefore, in the embodiment of the present application, a gray histogram is constructed to detect the target connected domain in each grayscale image, that is, the area of the moving objects such as vehicles and pedestrians.

[0055] Firstly, for any grayscale image, a gray histogram of the grayscale image is constructed, the horizontal axis of the gray histogram is the gray value, and the vertical axis is the number of pixel points corresponding to the gray value. The number of pixel points corresponding to the vertical axis in the gray histogram is connected to obtain a number change curve.

[0056] Then, considering that the grayish white of the heavy fog covers the pixel points in the moving object area, the number of pixel points in the moving object area is uneven, which is shown in the gray histogram as: the peak of the number change curve is relatively moderate, and the rising and falling trend on the left and right sides of the peak is relatively more gentle. Therefore, the peak points in the number change curve are obtained by using the AMPD peak finding algorithm, which belongs to the prior art and will not be described in detail here. Then, according to the local distribution characteristics of each peak point, the suspected target pixel points in the gray image are obtained, which are the pixel points belonging to the moving object.

[0057] The method for obtaining the suspected target pixel points in any gray image is:

[0058] For any peak point, the maximum pixel point number and the minimum pixel point number on the number change curve are obtained, the difference between the maximum pixel point number and the minimum pixel point number is calculated, which is denoted as the number range, the difference between the pixel point number corresponding to the peak point and the minimum pixel point number is calculated, which is denoted as the first difference, and the negative of the ratio between the first difference and the number range is taken as the independent variable of the preset exponential function, to obtain the pixel distribution ratio corresponding to the peak point.

[0059] The left valley point and the right valley point of the peak point in the number change curve are obtained, the curve between the left valley point and the right valley point is intercepted as the local sub-curve of the peak point, the slope between each coordinate point on the local sub-curve and its adjacent coordinate point is calculated respectively to obtain a slope sequence, the absolute value of the difference between each two adjacent slopes in the slope sequence is calculated to obtain an average difference absolute value, and the average difference absolute value is taken as the independent variable of the preset exponential function to obtain the local fluctuation characteristic value corresponding to the peak point.

[0060] The sum of the pixel distribution ratio and the local fluctuation characteristic value corresponding to the peak point is calculated to obtain the probability value of the peak point meeting the moving object characteristics. If the probability value of the peak point meeting the moving object characteristics is greater than or equal to a preset probability threshold, the gray value corresponding to the peak point is determined as a suspected target gray value.

[0061] All the suspected target gray values in the gray image are obtained, and the pixel points corresponding to all the suspected target gray values in the gray image are taken as the suspected target pixel points.

[0062] In an embodiment, taking the qth peak point on the quantity change curve as an example, the left adjacent valley point and the right adjacent valley point of the qth peak point on the quantity change curve are obtained, and the curve between the left adjacent valley point and the right adjacent valley point is taken as a local sub-curve of the qth peak point. In combination with the change of the local sub-curve, a calculation expression of a probability value that the qth peak point conforms to the moving object feature is obtained as follows:

[0063]

[0064] wherein Y q represents the probability value that the qth peak point conforms to the moving object feature, exp() represents an exponential function with a natural constant as a base number, p q represents the pixel point quantity corresponding to the qth peak point on the quantity change curve, that is, the ordinate value of the qth peak point, p min represents the minimum pixel point quantity in the quantity change curve, that is, the minimum ordinate value, p max represents the maximum pixel point quantity in the quantity change curve, that is, the maximum ordinate value, n q represents the coordinate point quantity on the local sub-curve of the qth peak point, k i represents the slope between the i th coordinate point and the i+1 th coordinate point on the local sub-curve of the qth peak point, k i+1 represents the slope between the i+1 th coordinate point and the i+2 th coordinate point on the local sub-curve of the qth peak point, and || represents an absolute value symbol.

[0065] It should be noted that, is used to represent the proportion of the pixel point quantity corresponding to the qth peak point in the overall pixel point quantity. The smaller the proportion value is, the greater the difference between the gray value corresponding to the qth peak point and other gray values is, and the greater the probability value that the gray value corresponding to the qth peak point conforms to the moving object feature is. is used to represent the fluctuation trend of the local sub-curve of the qth peak point, The smaller the value of is, the more relatively stable the rising or falling trend of the local sub-curve of the qth peak point is, the more in line with the characteristics of the moving object the qth peak point is, and the greater the probability value that the gray value corresponding to the qth peak point conforms to the moving object feature is.

[0066] The probability threshold is set to 0.8. If the probability value that the qth peak point conforms to the moving object feature is greater than or equal to 0.8, it is considered that the gray value corresponding to the qth peak point is a suspected target gray value, and the pixel point corresponding to the suspected target gray value is likely to be a pixel point in the moving object region. Therefore, the pixel point corresponding to the suspected target gray value is taken as a suspected target pixel point. Similarly, the probability values that all the peak points on the quantity change curve conform to the moving object feature are obtained to obtain all the suspected target pixel points in the gray image.

[0067] After all the suspected target pixel points in the gray image are determined, it is necessary to further obtain a suspected connected domain representing a moving object region according to the suspected target pixel points. The more similar the gray values of two pixel points are, the more likely they are in the same suspected connected domain (moving object region). Due to the influence of the fog environment, there may be some areas with relatively thick fog and relatively thin fog in the moving object region detected by vision, and the gray difference value obtained therefrom will be relatively large. However, the composition ratio of the RGB channel is not affected by the fog. When the fog is relatively thick, the RGB channel value will be reduced by the same proportion, and thus the color difference feature of the pixel point is used as another feature. When the color composition of two pixel points is more similar, that is, the color difference feature value is more similar, it is more likely that the two pixel points are in the same suspected connected domain. Therefore, in the embodiment of the present application, first, the color difference feature value of each pixel point in the gray image is obtained according to the RGB value of the corresponding pixel point, then the feature similarity degree between each suspected target pixel point and each pixel point in its preset neighborhood range is calculated according to the gray difference and the color difference feature value difference, and finally, at least one target connected domain in the gray image is obtained based on the feature similarity degree, which is the region corresponding to the suspected moving object.

[0068] Among them, the color difference feature value of each pixel point in the gray image is obtained according to the RGB value of the corresponding pixel point, including:

[0069] For any pixel point in any gray image, the absolute value of the difference between the R channel value and the G channel value, the absolute value of the difference between the G channel value and the B channel value, and the absolute value of the difference between the R channel value and the B channel value are calculated respectively according to the R channel value, the G channel value and the B channel value of the pixel point, the cumulative value of all the absolute values of the differences is obtained, and the cumulative value is normalized to obtain the color difference feature value of the pixel point.

[0070] In an embodiment, in the RGB image, the surrounding environment region in heavy fog weather is mostly grayish white, and the difference between the R value, the G value and the B value of the pixel point is small. However, the moving objects such as pedestrians and vehicles have more colors due to the clothes and the objects or vehicles they carry, and the difference between the R value, the G value and the B value is relatively larger. Therefore, taking the jth pixel point in the gray image as an example, the calculation expression of the color difference feature value of the jth pixel point is:

[0071] C j = norm [ | R j - G j | + | R j - B j | + | G j - B j | ]

[0072] wherein, C j represents the color difference feature value of the jth pixel point, norm() represents a normalization function, R j represents the R channel value of the jth pixel point, G j represents the G channel value of the jth pixel point, B j represents the B channel value of the jth pixel point, and || represents an absolute value symbol.

[0073] It should be noted that, |R p -G p |+|R p -B p |+|G p -B p |is the difference accumulation between different channel values, and the greater the difference accumulation, the more obvious the color difference of the jth pixel point is, and the more likely it is to be a pixel point in the moving object region in the heavy fog environment.

[0074] wherein, according to the gray difference and the color difference feature value difference, the feature similarity degree between each suspected target pixel point and each pixel point in the preset neighborhood range thereof is calculated, including:

[0075] any pixel point in the preset neighborhood range of the any suspected target pixel point is taken as a neighborhood pixel point, and the absolute value sum of the gray value difference and the difference feature value between the any suspected target pixel point and the neighborhood pixel point is calculated;

[0076] the addition value between the reciprocal of the absolute value sum of the gray value difference and the reciprocal of the absolute value sum of the difference feature value is calculated, to obtain the feature similarity degree between the any suspected target pixel point and the neighborhood pixel point.

[0077] In an embodiment, the preset neighborhood range is an eight-neighborhood, and taking the vth suspected target pixel point as an example, the calculation expression of the feature similarity degree between the vth suspected target pixel point and the uth pixel point in the eight-neighborhood thereof is:

[0078]

[0079] wherein, R v_u represents the feature similarity degree between the vth suspected target pixel point and the uth pixel point in the eight-neighborhood thereof, || represents an absolute value symbol, G v represents the gray value of the vth suspected target pixel point, G v represents the gray value of the uth pixel point in the eight-neighborhood of the vth suspected target pixel point, C v represents the color difference feature value of the vth suspected target pixel point, and C uA color difference feature value of a u-th pixel point in an eight-neighborhood of a v-th suspected target pixel point.

[0080] It should be noted that, the greater the value of |C v -G u The greater the value of |C v -C u The smaller the value of |C

[0081] The at least one target connected domain in the gray-scale image is obtained based on the feature similarity degree, and the method comprises the following steps:

[0082] For any suspected target pixel point in any gray-scale image, the feature similarity degree between the suspected target pixel point and each pixel point in the preset neighborhood range of the suspected target pixel point is calculated according to the gray-scale difference and the color difference feature value difference, and if there is any feature similarity degree greater than or equal to a preset feature similarity degree threshold, the pixel point corresponding to the any feature similarity degree is taken as a new suspected target pixel point.

[0083] The new suspected target pixel point is taken as the any suspected target pixel point, and the step of the any suspected target pixel point is repeated until all feature similarity degrees are less than the preset feature similarity degree threshold, so that all new suspected target pixel points are obtained.

[0084] All new suspected target pixel points corresponding to each suspected target pixel point in the any gray-scale image are obtained, and all suspected target pixel points and all new suspected target pixel points in the any gray-scale image are taken to form at least one target connected domain.

[0085] In an embodiment, the feature similarity threshold is set to 0.3. After obtaining the feature similarity between the vth suspected target pixel point and the u pixel point in its eight-neighborhood, if the feature similarity between the vth suspected target pixel point and the u pixel point in its eight-neighborhood is greater than or equal to 0.3, it is considered that the two pixel points are similar, and the vth suspected target pixel point and the u pixel point are connected to form a same connected domain, and the u pixel point is taken as a new suspected target pixel point. The eight-neighborhood of the u pixel point is traversed according to the eight-neighborhood traversal mode of the vth suspected target pixel point, and a new suspected target pixel point is continuously searched, and the process is repeated until there is no new suspected target pixel point in the eight-neighborhood of the pixel point, and all new suspected target pixel points corresponding to the vth suspected target pixel point are obtained. Similarly, all new suspected target pixel points corresponding to each suspected target pixel point are obtained, and then according to the positions of each suspected target pixel point and each new suspected target pixel point in the gray-scale image, at least one connected domain composed of suspected target pixel points and new suspected target pixel points is obtained, which is recorded as a target connected domain, that is, a region suspected to be a moving object such as a pedestrian or a vehicle.

[0086] After the car stops, if there is a moving object such as a pedestrian or a vehicle approaching gradually from the rear side, according to the visual effect feature that the near is large and the far is small, the moving object region will continue to increase, and in the continuous gray-scale images, the target connected domain of the same moving object will become larger and larger. Due to the covering effect of abnormal environment such as heavy fog, the gray-scale similarity range of the moving object region in the gray-scale image as a whole will decrease, that is, the overall uniformity will increase. As the moving object approaches gradually, the degree of hindering of heavy fog weather to visual detection will gradually decrease, and the overall gray-scale uniformity in the target connected domain will gradually decrease. Therefore, after all the target connected domains in each gray-scale image are determined, the possibility that the target connected domain belongs to the moving object region and the moving risk can be judged according to the multi-frame continuous change feature of the target connected domain.

[0087] Since the target connected domain in the gray-scale image is not only for one moving object, there can be multiple moving objects. In order to accurately analyze the moving risk of each moving object subsequently, the target connected domains belonging to the same moving object need to be divided first. Since the overlapping area of the same target connected domain between adjacent frames is usually large, in order to avoid multiple target connected domains with overlapping areas between adjacent frames, the target connected domains in all gray-scale images are divided according to the overlapping area between the target connected domains in adjacent gray-scale images, and multiple target connected domain sets are obtained, wherein the division method is:

[0088] For the mth target connected domain in the last gray image, the last gray image is obtained, the target connected domain with the maximum overlapping area with the mth target connected domain in the last gray image is obtained as the same target connected domain of the mth target connected domain;

[0089] The same target connected domain of the mth target connected domain is taken as the mth target connected domain in the last gray image, and the obtaining method of the same target connected domain of the mth target connected domain is repeated until the same target connected domain of the mth target connected domain is obtained in the first gray image, and the mth target connected domain and all the same target connected domains of the mth target connected domain are combined to form a mth target connected domain set.

[0090] In an embodiment, assuming that the last frame of gray images is the ath gray image, for the mth target connected domain in the ath gray image, the a-1th gray image is projected into the ath gray image, and then the target connected domain with an overlapping area with the mth target connected domain in the a-1th gray image is obtained as a candidate connected domain, and then the candidate connected domain corresponding to the maximum overlapping area is recorded as the same target connected domain of the mth target connected domain according to the overlapping area between each candidate connected domain and the mth target connected domain; then the a-2th gray image is projected into the a-1th gray image, and then the target connected domain with an overlapping area with the same target connected domain belonging to the mth target connected domain in the a-1th gray image in the a-2th gray image is obtained as a candidate connected domain, and then the candidate connected domain corresponding to the maximum overlapping area is recorded as the same target connected domain of the mth target connected domain according to the overlapping area between each candidate connected domain and the same target connected domain belonging to the mth target connected domain in the a-1th gray image, and the same target connected domain of the mth target connected domain is obtained in each gray image before the ath gray image, and the mth target connected domain and all the same target connected domains of the mth target connected domain are combined to form a mth target connected domain set. Similarly, a plurality of target connected domain sets are obtained.

[0091] Considering that one target connected domain set corresponds to one suspected moving object, and each moving object has a risk for automatic opening of the vehicle door, therefore, for any target connected domain set, the object moving risk degree corresponding to the any target connected domain set is obtained according to the area and gray distribution of the target connected domains of adjacent frames in the any target connected domain set, and the specific obtaining method is as follows:

[0092] respectively, and the absolute value of the difference between the area difference absolute value and the gray value variance absolute value is calculated, denoted as the inter-frame difference feature value between the two adjacent target connected domains in the any target connected domain set;

[0093] The inter-frame difference feature values between each two adjacent target connected domains in the any target connected domain set are accumulated to obtain the object movement risk degree corresponding to the any target connected domain set.

[0094] In an embodiment, taking the mth target connected domain set as an example, the calculation expression of the object movement risk degree corresponding to the mth target connected domain set is:

[0095]

[0096] Wherein, W m represents the object movement risk degree corresponding to the mth target connected domain set, S i-1 represents the area of the i-1th target connected domain in the mth target connected domain set, S i represents the area of the ith target connected domain in the mth target connected domain set, σ i-1 represents the gray value variance of all pixel points of the i-1th target connected domain in the mth target connected domain set, S i represents the gray value variance of all pixel points of the ith target connected domain in the mth target connected domain set, || represents the absolute value symbol, M m represents the number of target connected domains in the mth target connected domain set.

[0097] It should be noted that the greater the value of |S i-1 -S i , the faster the moving speed of the moving object corresponding to the mth target connected domain set, and the more obvious the movement change; the greater the value of |σ i-1 -σ i , the more the gray value change in the target connected domain of the adjacent frame in the mth target connected domain set has the trend from uniform to non-uniform, and the more it belongs to the movement characteristics of the moving object, therefore, the greater the value of W m , the more the mth target connected domain set belongs to the movement characteristics of the moving object, and the faster the moving speed, and the greater the corresponding object movement risk degree, and the higher the probability of danger of the automatic car door opening of the car.

[0098] Similarly, the object movement risk degree corresponding to each target connected domain set can be obtained. The greater the object movement risk degree, the greater the probability of danger when the automatic door of the vehicle is opened. Therefore, in the embodiment of the present application, the automatic door of the vehicle is automatically controlled according to the object movement risk degree corresponding to each target connected domain set, and the specific automatic control method is as follows:

[0099] A preset object movement risk degree threshold is obtained. If the object movement risk degree corresponding to any target connected domain set is greater than or equal to the object movement risk degree threshold, it is determined that a moving object is approaching, and the control signal of the automatic door of the vehicle is determined to be stopped. Otherwise, if the object movement risk degree corresponding to all target connected domain sets is less than the object movement risk degree threshold, the control signal of the automatic door of the vehicle is determined to be opened.

[0100] In an embodiment, the object movement risk degree corresponding to each target connected domain set is normalized to obtain a corresponding normalized value norm(W m ), and a normalized threshold of the object movement risk degree is set to 0.6. If the normalized value of the object movement risk degree corresponding to at least one target connected domain set is greater than or equal to 0.6, it is considered that the suspected moving object region corresponding to the target connected domain set has a movement risk, and the door cannot be opened at this time. The processor sends a stop control signal to the door control module. Otherwise, if the normalized value of the object movement risk degree corresponding to the target connected domain set is less than 0.6, the processor sends a continue opening control signal to the door control module.

[0101] After the processor obtains the control signal of the automatic door of the vehicle, the control signal is transmitted to the door control module.

[0102] The door control module is configured to control the automatic door of the vehicle to continue to open or stop according to the received control signal.

[0103] After the door control module receives the control signal, the door control module controls the automatic door of the vehicle to continue to open or stop according to the received control signal. Specifically, when the received control signal is stop, the door control module automatically locks the automatic door of the vehicle for a short time, and plays a voice prompt in the vehicle to remind the person in the vehicle to pay attention to the approach of the moving object such as the pedestrian and the vehicle behind, so as to avoid the danger caused by the manual opening of the person in the vehicle or the automatic opening of the automatic door. When the received control signal is open, the door control module controls the automatic door to continue to open until the door is opened to a preset angle.

[0104] The technical scheme of the present application solves the deficiencies of the existing automatic door system in identifying rear objects, especially moving objects, by combining pure vision detection technology. The system uses a high-precision camera on the vehicle to monitor and capture image information of static and dynamic objects located on the side and rear of the door frame in real time. Through its own core algorithm, the system performs multi-level enhancement processing on the collected image data to improve the clarity and resolution of the images, ensuring that the system can still operate stably even under complex lighting or weather conditions.

[0105] During image processing, the system can accurately identify the specific positions of static and moving objects and predict their dynamic behavior through motion tracking technology. By real-time estimation of dynamic parameters such as object direction and speed, the system can determine the relative position of the object to the door and the collision risk at an early stage. Based on this determination, the system can automatically assess the likelihood of a collision and issue a warning signal in time to alert the driver or vehicle owner when the risk reaches a certain threshold.

[0106] More importantly, the technical scheme of the present application seamlessly integrates this intelligent identification and evaluation system with the automatic door control system. Once the collision risk is detected, the system can intervene in the opening and closing actions of the door, automatically adjusting the opening speed or directly stopping the door from opening, thereby avoiding collisions with rear objects, especially moving objects. This intelligent linkage mechanism not only improves the safety of automatic door operation but also enhances the overall response speed and accuracy of the vehicle's intelligent system.

[0107] Through this innovative solution, the present application effectively solves the blind spot problem of current automatic door systems in identifying rear objects, especially moving objects, especially dynamic obstacles that are easily overlooked during door opening, significantly reducing the risk of collision accidents. The implementation of this technical scheme not only improves the intelligence level of the automatic door system but also provides a safer and more convenient user experience for vehicles in practical applications, further promoting the application and popularization of intelligent transportation technology in daily driving.

[0108] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A purely visual-based automobile automatic door anti-collision system, characterized in that: The system comprises: At least one camera, disposed on a door frame of the vehicle, for collecting images of the environment behind and to the sides of the vehicle when the automatic door of the vehicle automatically opens at a preset distance; a processor, connected to the camera and the door control module, configured to receive an environmental image captured by the camera, detect and identify moving objects behind and to the side of the vehicle based on the environmental image, obtain a control signal for the vehicle's automatic door, and send the control signal to the door control module; The door control module is used to control the car's automatic door to continue opening or stopping according to the received control signal; The detecting and identifying moving objects behind and to the side of the vehicle according to the environmental image to obtain a control signal for the vehicle's automatic door includes: Acquire at least two frames of environmental images captured by any camera, and perform grayscale processing on each frame of the environmental image to obtain a corresponding grayscale image; For any grayscale image, construct a grayscale histogram of the grayscale image, and obtain a target connected domain in the grayscale image based on the grayscale histogram and the RGB value of each pixel in the grayscale image, wherein the target connected domain refers to a region of moving objects including vehicles and pedestrians; Obtaining a target connected domain in each of the grayscale images, dividing the target connected domains in all grayscale images according to overlapping areas between target connected domains in adjacent grayscale images to obtain a plurality of target connected domain sets, and obtaining, for any target connected domain set, a degree of object movement risk corresponding to the target connected domain set according to the area and grayscale distribution of target connected domains in adjacent frames in the target connected domain set; A control signal for the automatic door of the car is obtained according to the object movement risk degree corresponding to each target connected domain set.

2. The purely visual-based automobile automatic door anti-collision system according to claim 1, characterized in that: The step of obtaining a target connected domain in any grayscale image according to the grayscale histogram and the RGB value of each pixel in the any grayscale image includes: Connecting the numbers of pixels corresponding to the vertical axis in the grayscale histogram to obtain a number change curve, obtaining peak points in the number change curve, and obtaining suspected target pixel points in any grayscale image based on the local distribution characteristics of each peak point; According to the RGB value of each pixel in any grayscale image, the color difference feature value of the corresponding pixel is obtained, and according to the grayscale difference and color difference feature value of each suspected target pixel and each pixel in its local neighborhood, the target connected domain in any grayscale image is obtained.

3. The purely visual-based automobile automatic door anti-collision system according to claim 2, characterized in that: The obtaining of the suspected target pixel point in any grayscale image according to the local distribution characteristics of each peak point includes: For any peak point, obtain the maximum number of pixels and the minimum number of pixels on the number change curve, calculate the difference between the maximum number of pixels and the minimum number of pixels, record it as the number range, calculate the difference between the number of pixels corresponding to any peak point and the minimum number of pixels, record it as the first difference, use the negative of the ratio between the first difference and the number range as the independent variable of a preset exponential function, and obtain the pixel distribution ratio corresponding to any peak point; Obtaining a left valley point and a right valley point of any peak point in the quantity change curve, intercepting a curve between the left valley point and the right valley point as a local subcurve of the any peak point, respectively calculating the slope between each coordinate point and its subsequent adjacent coordinate point on the local subcurve to obtain a slope sequence, calculating the absolute value of the difference between every two adjacent slopes in the slope sequence to obtain an average absolute value of the difference, using the average absolute value of the difference as an independent variable of a preset exponential function, and obtaining a local fluctuation characteristic value corresponding to the any peak point; Calculate the sum of the pixel distribution ratio and the local fluctuation characteristic value corresponding to any peak point to obtain a probability value that any peak point meets the characteristics of a moving object. If the probability value that any peak point meets the characteristics of a moving object is greater than or equal to a preset probability threshold, determine that the grayscale value corresponding to the any peak point is a suspected target grayscale value; All suspected target grayscale values ​​in any grayscale image are obtained, and pixel points corresponding to all suspected target grayscale values ​​in any grayscale image are used as suspected target pixel points.

4. The purely visual-based automobile automatic door anti-collision system according to claim 2, characterized in that: The step of obtaining a color difference feature value of a corresponding pixel point according to the RGB value of each pixel point in any grayscale image includes: For any pixel point in any grayscale image, according to the R channel value, G channel value and B channel value of any pixel point, the absolute value of the difference between the R channel value and the G channel value, the absolute value of the difference between the G channel value and the B channel value, and the absolute value of the difference between the R channel value and the B channel value are calculated respectively, and the accumulated value of all the absolute values ​​of the difference is obtained accordingly. The accumulated value is normalized to obtain the color difference characteristic value of any pixel point.

5. The purely visual-based automobile automatic door anti-collision system according to claim 2, characterized in that: The step of obtaining the target connected domain in any grayscale image based on the grayscale difference and color difference feature values ​​of each suspected target pixel and each pixel in its local neighborhood includes: For any suspected target pixel in any grayscale image, calculate the feature similarity between the suspected target pixel and each pixel in its preset neighborhood based on the grayscale difference and color difference feature value differences. If any feature similarity is greater than or equal to a preset feature similarity threshold, use the pixel corresponding to the feature similarity as a new suspected target pixel. The new suspected target pixel point is used as any suspected target pixel point, and the steps of any suspected target pixel point are repeated until all feature similarities are less than a preset feature similarity threshold, thereby obtaining all new suspected target pixel points; All new suspected target pixels corresponding to each suspected target pixel in any grayscale image are obtained, and all suspected target pixels in any grayscale image and all new suspected target pixels are combined to form at least one target connected domain.

6. The purely visual-based automobile automatic door anti-collision system according to claim 5, characterized in that: Calculating the feature similarity between any suspected target pixel and each pixel within its preset neighborhood based on the grayscale difference and color difference feature value difference includes: Taking any pixel point within a preset neighborhood range of any suspected target pixel point as a neighborhood pixel point, calculating the absolute value of the grayscale value difference and the absolute value of the color difference feature value difference between any suspected target pixel point and the neighborhood pixel point; The sum of the inverse of the absolute value of the grayscale value difference and the inverse of the absolute value of the color difference feature value difference is calculated to obtain the feature similarity between any suspected target pixel and the neighborhood pixel.

7. The purely visual-based automobile automatic door anti-collision system according to claim 1, characterized in that: The target connected domains in all grayscale images are divided according to the overlapping areas between the target connected domains in adjacent grayscale images to obtain multiple target connected domain sets, including: For the mth target connected domain in the last grayscale image, obtain the previous grayscale image of the last grayscale image, and obtain the target connected domain with the largest overlapping area with the mth target connected domain in the previous grayscale image, and record it as the same target connected domain of the mth target connected domain; The same target connected domain of the m-th target connected domain is used as the m-th target connected domain in the last grayscale image, and the method of obtaining the same target connected domain of the m-th target connected domain is repeated until the same target connected domain of the m-th target connected domain is obtained in the first grayscale image, and the m-th target connected domain and the same target connected domain of all m-th target connected domains are combined into an m-th target connected domain set.

8. The purely visual-based automobile automatic door anti-collision system according to claim 7, characterized in that: The obtaining, based on the area and grayscale distribution of the target connected domains of adjacent frames in any target connected domain set, the object movement risk level corresponding to any target connected domain set includes: Obtain the variance of the grayscale values ​​of all pixels in each target connected domain in any target connected domain set, respectively, and record it as the grayscale value variance of the corresponding target connected domain; for any two adjacent target connected domains in any target connected domain set, obtain the absolute value of the area difference and the absolute value of the grayscale value variance difference between the two adjacent target connected domains, calculate the sum of the absolute value of the area difference and the absolute value of the grayscale value variance difference, and record it as the inter-frame difference feature value between the two adjacent target connected domains; The inter-frame difference feature values ​​between every two adjacent target connected domains in any target connected domain set are accumulated to obtain the object movement risk degree corresponding to any target connected domain set.

9. The purely visual-based automobile automatic door anti-collision system according to claim 1, characterized in that: The step of obtaining a control signal for an automatic door of a car according to the object movement risk level corresponding to each target connected domain set includes: Obtain a preset object movement risk level threshold. If the object movement risk level corresponding to any target connected domain set is greater than or equal to the object movement risk level threshold, determine that the control signal of the car automatic door is stopped; if the object movement risk levels corresponding to all target connected domain sets are less than the object movement risk level threshold, determine that the control signal of the car automatic door is opened.

Images