Parallax-based ground reflection misrecognition detection method, system and electronic equipment

Abstract

The present application discloses a method, system and electronic device for detecting misidentification of ground reflections based on parallax, wherein the method includes: obtaining a target frame in a left image based on a deep learning model detection, calculating a first distance value and a second distance value using monocular ranging and binocular ranging respectively; calculating the distance difference between the first distance value and the second distance value, and when the distance difference is greater than a difference threshold, calculating the disparity distribution interval within the target frame range in the disparity map; performing regional growth in the disparity map based on the target frame and the disparity distribution interval to determine the disparity growth frame; when the difference between the target frame and the disparity growth frame in the height direction is greater than the height threshold, determining that the target to be identified corresponding to the target frame is a ground reflection. Through the technical solution in the present application, it is possible to effectively detect misidentification of ground reflections based on parallax, which helps to ensure the driving safety of autonomous driving vehicles.

Description

Technical Field

[0001] The present application relates to the technical field of binocular vision, and in particular to a parallax-based ground reflection misrecognition detection method, a parallax-based ground reflection misrecognition detection system, and an electronic device. Background Art

[0002] In pure vision-based autonomous driving technology, AI models can be used to identify the type of obstacles in front of the vehicle, and binocular vision technology can be used to obtain the distance (depth) between the obstacle and the vehicle, so as to perceive the scene in front of the vehicle.

[0003] However, in some scenarios, such as water on the road surface, the AI ​​model will identify the reflective area in the image as an obstacle, and the disparity map will also contain the disparity information of such obstacles, which will lead to misidentification of obstacle detection and trigger false alarms for autonomous vehicles. Figure 1 As shown, due to water accumulation on the road, the taillights of the vehicle ahead will form a reflective area on the road surface (such as target box 101), which may be identified as a "triangular cone", "ice cream cone", etc. Such misidentification is extremely difficult to completely avoid through AI training alone.

[0004] In the prior art, such as the Chinese invention patent application "Road Reflective Area Detection Method, Device and Terminal" with publication number CN108573215A, the road surface plane is usually fitted in the disparity map, and then based on the positional relationship between the detected obstacle and the fitted road surface plane, it is determined whether the obstacle is a real obstacle.

[0005] Those skilled in the art will understand that the reliability of such methods is entirely dependent on the road surface plane fitted from the disparity map, while the accuracy of the fitted road surface plane itself is not high. This leads to some inevitable false detections when performing threshold screening based on the road surface plane, especially for smaller reflective areas in some images. Summary of the invention

[0006] The purpose of this application is to detect misidentification of ground reflections to ensure the driving safety of autonomous vehicles.

[0007] The technical solution of the first aspect of the present application is: a parallax-based ground reflection misidentification detection method is provided, the method comprising: obtaining a target frame in a left image based on a deep learning model detection, and calculating a first distance value and a second distance value using monocular ranging and binocular ranging respectively; calculating the distance difference between the first distance value and the second distance value, and when the distance difference is greater than a difference threshold, calculating the parallax distribution interval within the target frame range in the disparity map; performing region growing in the disparity map based on the target frame and the parallax distribution interval to determine a parallax growth frame; when the difference between the target frame and the parallax growth frame in the height direction is greater than a height threshold, determining that the target to be identified corresponding to the target frame is a ground reflection.

[0008] In any of the above technical solutions, further, the disparity distribution interval within the target frame range of the disparity map is calculated, specifically including: traversing the disparity value of any disparity point in the disparity map within the target frame range to generate a disparity projection map; normalizing and binarizing the disparity projection map to generate a binary projection map; based on the non-zero maximum value and non-zero minimum value in the target axis of the binary projection map, calculating the disparity distribution interval, wherein, when the disparity projection map is a disparity U map, the target axis is the vertical axis, and when the disparity projection map is a disparity V map, the target axis is the horizontal axis.

[0009] In any of the above technical solutions, the method further includes: scaling the disparity distribution interval according to a preset ratio; performing region growing in the disparity map based on the target frame and the scaled disparity distribution interval to determine the disparity growth frame.

[0010] In any of the above technical solutions, further, based on the target frame and the disparity distribution interval, regional growth is performed in the disparity map to determine the disparity growth frame, specifically including: searching for unmarked disparity points in the disparity map within the range of the target frame in order from left to right and from top to bottom; taking any unmarked disparity point as the origin, determining whether the disparity value of the disparity point within the search radius is within the disparity distribution interval, if so, marking the disparity point as the first disparity point, if not, marking the disparity point as the second disparity point; determining the disparity growth frame based on the minimum circumscribed rectangle of the first disparity point.

[0011] In any of the above technical solutions, further, the value of the search radius is determined by the median value of the disparity of the disparity points within the target frame.

[0012] The technical solution of the second aspect of the present application is: a parallax-based ground reflection misidentification detection system is provided, the system comprising: a distance calculation unit, the distance calculation unit is configured to obtain a target frame in a left image based on a deep learning model detection, and calculate a first distance value and a second distance value using monocular ranging and binocular ranging respectively; an interval determination unit, the interval determination unit is configured to calculate a distance difference between the first distance value and the second distance value, and when the distance difference is greater than a difference threshold, calculate a parallax distribution interval within the target frame range in the parallax map; a growth frame determination unit, the growth frame determination unit is configured to perform regional growth in the parallax map based on the target frame and the parallax distribution interval, and determine a parallax growth frame; a determination unit, the determination unit is configured to determine that a target to be identified corresponding to the target frame is a ground reflection when the difference between the target frame and the parallax growth frame in the height direction is greater than a height threshold.

[0013] In any of the above technical solutions, further, the disparity distribution interval within the target frame range of the disparity map is calculated, specifically including: traversing the disparity value of any disparity point in the disparity map within the target frame range to generate a disparity projection map; normalizing and binarizing the disparity projection map to generate a binary projection map; based on the non-zero maximum value and non-zero minimum value in the target axis of the binary projection map, calculating the disparity distribution interval, wherein, when the disparity projection map is a disparity U map, the target axis is the vertical axis, and when the disparity projection map is a disparity V map, the target axis is the horizontal axis.

[0014] In any of the above technical solutions, the system further includes: a scaling unit, which is configured to scale the disparity distribution interval according to a preset ratio; a growth frame determination unit, which is also configured to: perform regional growth in the disparity map based on the target frame and the scaled disparity distribution interval to determine the disparity growth frame.

[0015] In any of the above technical solutions, further, the growth frame determination unit is specifically configured as follows: searching for unmarked disparity points in the disparity map within the range of the target frame in order from left to right and from top to bottom; taking any unmarked disparity point as the origin, determining whether the disparity value of the disparity point within the search radius is within the disparity distribution interval, if so, marking the disparity point as a first disparity point, if not, marking the disparity point as a second disparity point; and determining the disparity growth frame based on the minimum circumscribed rectangle of the first disparity point.

[0016] The technical solution of the third aspect of the present application is: an electronic device is provided, including: a memory, the memory stores execution instructions; a processor, the processor executes the execution instructions stored in the memory, so that the processor executes the parallax-based ground reflection misidentification detection method of any one of the above-mentioned first aspect technical solutions.

[0017] The beneficial effects of this application are:

[0018] The technical solution in this application uses two different distance measurement methods to obtain the distance value of the target detection frame, and uses the difference between the two as the basis for judging whether it is a potential misidentified target. After that, in the disparity map (disparity U map or disparity V map), the disparity distribution interval of such potential misidentified targets is calculated, and regional growth is performed to obtain the disparity growth frame corresponding to the target frame, and then the difference between the two frames in the height direction is used as the basis for judging whether it is a misidentification. By using 3D physical rules, the problem of misidentification of deep learning models is solved, and the misidentification of ground reflections can be effectively detected based on parallax, which helps to ensure the driving safety of autonomous vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The advantages of the above and / or additional aspects of the present application will become apparent and easily understood in the description of the embodiments in conjunction with the following drawings, in which:

[0020] Figure 1 It is a schematic diagram of misidentification of road surface reflective areas in the prior art;

[0021] Figure 2 is a schematic flow chart of a parallax-based ground reflection misidentification detection method according to an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of a disparity map including a road surface reflective area according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of a disparity growth frame in a disparity map according to an embodiment of the present application;

[0024] Figure 5 is a schematic diagram of a disparity V map according to an embodiment of the present application;

[0025] Figure 6 It is a schematic block diagram of a parallax-based ground reflection misidentification detection system according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0027] In the following description, many specific details are elaborated to facilitate a full understanding of the present application. However, the present application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present application is not limited to the specific embodiments disclosed below.

[0028] Embodiment 1:

[0029] like Figure 2 As shown, this embodiment provides a method for detecting misidentification of ground reflections based on parallax, the method comprising:

[0030] Step 1: Detect the target frame in the left image based on the deep learning model, and calculate the first distance value and the second distance value using monocular ranging and binocular ranging respectively;

[0031] Specifically, after the binocular camera is installed on the vehicle and calibrated, the left image and the right image within the front field of view can be obtained; then based on the principle of parallax calculation, the disparity map corresponding to the left image can be obtained.

[0032] for Figure 1 For the scene in FIG. 1 , by using a deep learning model to perform target recognition on the left image captured by the binocular camera, the “triangular cone” (target box 102) generated by the front vehicle (target box 101) and the reflection of the taillight can be obtained.

[0033] In the corresponding disparity map, Figure 3 As shown, both target frames have corresponding parallax areas, so through binocular ranging, the objects in these two target frames will be given "distance" information. However, for target frame 101, it is a misdetected obstacle, and if it is given "distance" information, it will trigger a false alarm of the autonomous driving.

[0034] Considering that the binocular camera can obtain the left image and the right image, which are equivalent to two monocular images, a monocular ranging method can be introduced to measure the distance of the targets in the above two target frames. Among them, the monocular ranging method can choose to estimate the distance based on the image scale, that is, the size of the target category is taken as known information, and then according to the perspective principle, the width or height of the image pixels is used to estimate the distance of the target.

[0035] In this embodiment, the monocular distance measurement result is used as the first distance value, and the binocular distance measurement result is used as the second distance value.

[0036] Step 2: Calculate the distance difference between the first distance value and the second distance value. When the distance difference is greater than the difference threshold, calculate the disparity distribution interval within the target frame range in the disparity map.

[0037] Considering that the above-mentioned monocular ranging method calculates the first distance based on the pre-calibrated road surface reference and target size, and the bottom position of the "pseudo target" formed by reflection is lower than the real target, and the above-mentioned road surface reference itself also has deviations, therefore, there will be a difference between the above-mentioned first distance value and the second distance value, wherein, for the target frame 101, the first distance value obtained by monocular ranging is smaller than the second distance value obtained by binocular ranging.

[0038] For the vehicle in the target frame 102 , since its bottom is located on the road reference, even if there is a distance difference (difference) between the first and second distances, this distance difference is smaller than the distance difference corresponding to the target frame 101 .

[0039] Based on this, a difference threshold is set in this embodiment. When the distance difference between the first and second distances corresponding to any target frame is greater than the difference threshold, the target frame is considered to be a "pseudo target" formed by reflection. If the distance difference is less than or equal to the difference threshold, the corresponding distance difference is considered to be a difference caused by the algorithm itself and can be ignored.

[0040] Since the disparity map corresponds to the left image one by one, the target frame 101 and the target frame 102 have corresponding disparity areas in the disparity map, such as Figure 3 As shown in the middle frame 301 and frame 302. For these two disparity regions, the disparity distribution intervals corresponding to the two target frames can be obtained by disparity value statistics.

[0041] It should be noted that the disparity distribution intervals corresponding to the two target frames may be the same or different, which is determined by the accuracy of the specific disparity calculation.

[0042] Step 3: Based on the target frame and the disparity distribution interval, perform region growth in the disparity map to determine the disparity growth frame;

[0043] Step 4: When the height difference between the target frame and the parallax growth frame is greater than the height threshold, it is determined that the target to be identified corresponding to the target frame is a ground reflection.

[0044] Now take frame 301 as an example for explanation. Take any disparity point with a disparity value in the frame 301 as a seed point, and determine whether the disparity value of the disparity point adjacent to it or with a step size less than a certain threshold is within the disparity distribution interval. If it is within the interval, then the disparity point is determined to meet the regional growth condition and is marked as a point within the growth area; if it is outside the interval, then the disparity point does not meet the regional growth condition and is marked as a point outside the growth area, or it is not marked and is not used as a seed point for the next regional growth.

[0045] After one region growing is completed, the next unmarked disparity point is found and the next region growing is performed until there is no disparity point satisfying the region growing condition in the region corresponding to the frame 301 .

[0046] Finally, based on the points in the growth area marked during the growth process, the corresponding parallax growth frame is obtained, such as Figure 4 As shown, the parallax growth box can be the minimum circumscribed rectangle of these points.

[0047] It should be noted that the parallax growth frame of frame 301 and the parallax growth frame of frame 302 may be the same.

[0048] Taking into account the false detection caused by the reflective area, its size must be smaller than the size of the target causing the reflection, that is, the size of the "triangular cone" frame 301 in the disparity map should be smaller than the size of the "vehicle" frame 302.

[0049] Therefore, for such misdetected targets, the difference between the target frame corresponding to the “triangular cone” and the parallax growth frame in the height direction is greater than the difference between the target frame corresponding to the “vehicle” and the parallax growth frame in the height direction.

[0050] At the same time, considering that there may be errors in the calculation process of the disparity map and the determination process of the target frame, even for normal targets, there may be differences between the target frame and the disparity growth frame in the height direction.

[0051] Therefore, by setting a height threshold, when the difference between the target frame and the parallax growth frame in the height direction is greater than the height threshold, the target to be identified corresponding to the target frame is determined to be a ground reflection; otherwise, it is considered to be a normal target.

[0052] In any of the above embodiments, further, in order to improve the calculation efficiency and accuracy of the disparity distribution interval, the disparity map may be projected and statistics may be performed in the disparity U map or the disparity V map. Therefore, calculating the disparity distribution interval within the target frame range in the disparity map specifically includes:

[0053] Traverse the disparity value of any disparity point in the disparity map within the target frame range to generate a disparity projection map;

[0054] Normalizing and binarizing the disparity projection image to generate a binary projection image;

[0055] The disparity distribution interval is calculated based on the non-zero maximum value and the non-zero minimum value in the target axis of the binary projection image, wherein when the disparity projection image is a disparity U image, the target axis is the vertical axis, and when the disparity projection image is a disparity V image, the target axis is the horizontal axis.

[0056] like Figure 5 As shown, the disparity V map is taken as an example for explanation, in which 501 is the disparity V map area corresponding to the vehicle, 502 is the disparity V map area corresponding to the "triangular cone" generated by the reflection, and 503 is the fitted ground straight line.

[0057] The target frame is mapped to the disparity map, and the disparity value of each disparity point in the disparity map within the target frame is traversed. The size of the disparity space / disparity value is used as the horizontal coordinate and the height of the image is used as the vertical coordinate to generate a disparity projection map in the disparity V map within the target frame.

[0058] The parallax projection image is normalized to obtain an 8-bit grayscale image, and the normalized value range is set to 0-255. Then, the grayscale image is binarized, wherein the binarization threshold can be set to 127, or can be adjusted according to actual conditions.

[0059] Finally, the horizontal axis is used as the target axis to search on the generated binary image to obtain the minimum and maximum non-zero horizontal axis disparity values. These two values ​​constitute the disparity distribution interval.

[0060] In any of the above embodiments, further, considering that in a sparse disparity map, for an object with a blurred boundary and background, the disparity map may not necessarily contain disparity points at its boundary, therefore, in order to obtain a disparity growth frame as accurately as possible, the method further includes:

[0061] According to a preset ratio, the parallax distribution interval is scaled, for example, it is enlarged to 1.1 times;

[0062] Based on the target frame and the scaled disparity distribution interval, region growing is performed in the disparity map to determine the disparity growth frame.

[0063] In any of the above embodiments, further, based on the target frame and the disparity distribution interval, performing region growing in the disparity map to determine the disparity growth frame specifically includes:

[0064] Search for unmarked disparity points in the disparity map within the target frame in order from left to right and from top to bottom;

[0065] By traversing, any unmarked disparity point is used as the origin / seed point, and region growing is performed in the entire disparity map to determine whether the disparity value of the disparity point within the search radius is within the disparity distribution interval.

[0066] Specifically, in the process of region growing, first, any unmarked disparity point within the range of the target frame is taken as the origin / seed point, and it is determined whether the disparity value of the disparity point within the range of a circle with it as the center and the search radius as the radius is within the disparity distribution interval. If it is within the disparity distribution interval, it is considered that the disparity point meets the region growing conditions; then, it is taken as the starting point (center of the circle) for the next growth, and a circle with the same radius is re-determined to find disparity points that meet the region growing conditions until there are no disparity points that meet the region growing conditions in the circle; finally, an unmarked disparity point is searched again within the range of the target frame, and it is used as the origin / seed point, and the above process is repeated until all disparity points within the range of the target frame are traversed.

[0067] It should be noted that the region growing process is performed in the entire disparity map, and during the search, it is only necessary to ensure that the disparity value is within the disparity distribution interval, and there is no need to require that the position coordinates of the searched disparity points are continuous.

[0068] If yes, the disparity point is marked as the first disparity point; if no, the disparity point is marked as the second disparity point, and such marked second disparity point will not be searched again in the next traversal to search for the origin / seed point. It should be noted that the second disparity point can also be directly filtered out.

[0069] A disparity growth box is determined based on the minimum bounding rectangle of the first disparity point.

[0070] In any of the above embodiments, further, the value of the search radius is determined by the median value of the disparity of the disparity points within the target frame, and the corresponding calculation formula is:

[0071] R=f(disparity)=α*disparity

[0072] Where R is the search radius, α is the preset scale factor, which is used to represent the conversion coefficient between the real physical scale and the image scale. It can be set according to actual needs, and disparity is the median disparity of the disparity points within the target frame.

[0073] Embodiment 2:

[0074] like Figure 6 As shown, this embodiment provides a parallax-based ground reflection misrecognition detection system 100, which includes:

[0075] A distance calculation unit 11 is configured to detect a target frame in the left image based on a deep learning model, and calculate a first distance value and a second distance value using monocular distance measurement and binocular distance measurement respectively;

[0076] An interval determination unit 12, the interval determination unit 12 is configured to calculate a distance difference between the first distance value and the second distance value, and when the distance difference is greater than a difference threshold, calculate a disparity distribution interval within a target frame range in the disparity map;

[0077] A growth frame determining unit 13, the growth frame determining unit 13 is configured to perform region growing in the disparity map based on the target frame and the disparity distribution interval to determine the disparity growth frame;

[0078] The determination unit 14 is configured to determine that the target to be identified corresponding to the target frame is a ground reflection when the difference between the target frame and the parallax growth frame in the height direction is greater than a height threshold.

[0079] In any of the above embodiments, further, calculating the disparity distribution interval within the target frame range in the disparity map, specifically including: traversing the disparity value of any disparity point in the disparity map within the target frame range to generate a disparity projection map; normalizing and binarizing the disparity projection map to generate a binary projection map; calculating the disparity distribution interval based on the non-zero maximum value and non-zero minimum value in the target axis of the binary projection map, wherein, when the disparity projection map is a disparity U map, the target axis is the vertical axis, and when the disparity projection map is a disparity V map, the target axis is the horizontal axis.

[0080] In any of the above embodiments, further, the system 100 further includes:

[0081] A scaling unit 15, the scaling unit 15 is configured to scale the parallax distribution interval according to a preset ratio;

[0082] The growth frame determination unit 13 is further configured to:

[0083] Based on the target frame and the scaled disparity distribution interval, region growing is performed in the disparity map to determine the disparity growth frame.

[0084] In any of the above embodiments, further, the growth frame determination unit 13 is specifically configured to: search for unmarked disparity points in the disparity map within the range of the target frame in an order from left to right and from top to bottom; take any unmarked disparity point as the origin, determine whether the disparity value of the disparity point within the search radius is within the disparity distribution interval, if so, mark the disparity point as a first disparity point, if not, mark the disparity point as a second disparity point; determine the disparity growth frame based on the minimum circumscribed rectangle of the first disparity point.

[0085] In any of the above embodiments, further, the value of the search radius is determined by the median value of the disparity of the disparity points within the target frame, and the corresponding calculation formula is:

[0086] R=f(disparity)=α*disparity

[0087] Where R is the search radius, α is the preset scale factor, which is used to represent the conversion coefficient between the real physical scale and the image scale. It can be set according to actual needs, and disparity is the median disparity of the disparity points within the target frame.

[0088] Embodiment 3:

[0089] This embodiment also provides an electronic device, including: a memory, the memory storing execution instructions; and a processor or other hardware module, the processor or other hardware module executes the execution instructions stored in the memory, so that the processor or other hardware module executes the above method.

[0090] So far, the various embodiments of the present application have been described in detail. In order to avoid obscuring the concept of the present application, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solution disclosed herein.

[0091] Although some specific embodiments of the present application have been described in detail through examples, those skilled in the art should understand that the above examples are only for illustration rather than for limiting the scope of the present application.

[0092] The steps in this application can be adjusted in order, combined, and deleted according to actual needs.

[0093] Although the present application is disclosed in detail with reference to the accompanying drawings, it should be understood that these descriptions are merely exemplary and are not intended to limit the application of the present application. The scope of protection of the present application is defined by the appended claims and may include various modifications, alterations and equivalents made to the invention without departing from the scope and spirit of the present application.

Claims

1. A parallax-based ground reflection misrecognition detection method, characterized in that: The method comprises: The target frame in the left image is detected based on the deep learning model, and the first distance value and the second distance value are calculated by using monocular ranging and binocular ranging respectively; Calculating a distance difference between the first distance value and the second distance value, and when the distance difference is greater than a difference threshold, calculating a disparity distribution interval within the target frame range in the disparity map; Based on the target frame and the disparity distribution interval, performing region growing in the disparity map to determine a disparity growth frame; When the difference between the target frame and the parallax growth frame in the height direction is greater than a height threshold, it is determined that the target to be identified corresponding to the target frame is a ground reflection.

2. The parallax-based ground reflection misrecognition detection method according to claim 1, characterized in that: The calculating of the disparity distribution interval within the target frame range in the disparity map specifically includes: Traversing the disparity value of any disparity point in the disparity map within the target frame range to generate a disparity projection map; Normalizing and binarizing the parallax projection image to generate a binary projection image; Calculating the disparity distribution interval based on the non-zero maximum value and the non-zero minimum value in the target axis of the binary projection image, Wherein, when the parallax projection image is a parallax U image, the target axis is the ordinate axis, When the parallax projection image is a parallax V image, the target axis is the horizontal coordinate axis.

3. The parallax-based ground reflection misrecognition detection method according to claim 1 or 2, characterized in that: The method further comprises: Scaling the disparity distribution interval according to a preset ratio; Based on the target frame and the scaled disparity distribution interval, region growing is performed in the disparity map to determine the disparity growth frame.

4. The parallax-based ground reflection misrecognition detection method according to claim 1, characterized in that: The performing region growing in the disparity map based on the target frame and the disparity distribution interval to determine the disparity growth frame specifically includes: Searching for unmarked disparity points in the disparity map within the range of the target frame in a left-to-right and top-to-bottom order; Taking any unmarked disparity point as the origin, determining whether the disparity value of the disparity point within the search radius is within the disparity distribution interval, If yes, mark the disparity point as a first disparity point, if no, mark the disparity point as a second disparity point; The disparity growth frame is determined based on a minimum circumscribed rectangle of the first disparity point.

5. The parallax-based ground reflection misrecognition detection method according to claim 4, characterized in that: The value of the search radius is determined by the median value of the disparity of the disparity points within the target frame.

6. A parallax-based ground reflection misrecognition detection system, characterized in that: The system comprises: A distance calculation unit, wherein the distance calculation unit is configured to detect a target frame in the left image based on the deep learning model, and calculate a first distance value and a second distance value by using monocular distance measurement and binocular distance measurement respectively; an interval determining unit, the interval determining unit being configured to calculate a distance difference between the first distance value and the second distance value, and when the distance difference is greater than a difference threshold, calculate a disparity distribution interval within the target frame range in the disparity map; a growth frame determining unit, the growth frame determining unit being configured to perform region growing in the disparity map based on the target frame and the disparity distribution interval to determine a disparity growth frame; A determination unit, wherein the determination unit is configured to determine that the target to be identified corresponding to the target frame is a ground reflection when the difference between the target frame and the parallax growth frame in the height direction is greater than a height threshold.

7. The parallax-based ground reflection misidentification detection system according to claim 6, characterized in that: The calculating of the disparity distribution interval within the target frame range in the disparity map specifically includes: Traversing the disparity value of any disparity point in the disparity map within the target frame range to generate a disparity projection map; Normalizing and binarizing the parallax projection image to generate a binary projection image; Calculating the disparity distribution interval based on the non-zero maximum value and the non-zero minimum value in the target axis of the binary projection image, Wherein, when the parallax projection image is a parallax U image, the target axis is the ordinate axis, When the parallax projection image is a parallax V image, the target axis is the horizontal coordinate axis.

8. The parallax-based ground reflection misidentification detection system according to claim 6 or 7, characterized in that: The system further comprises: A scaling unit, wherein the scaling unit is configured to scale the parallax distribution interval according to a preset ratio; The growth frame determination unit is further configured to: Based on the target frame and the scaled disparity distribution interval, region growing is performed in the disparity map to determine the disparity growth frame.

9. The parallax-based ground reflection misidentification detection system according to claim 6, characterized in that: The growth frame determination unit is specifically configured as follows: Searching for unmarked disparity points in the disparity map within the range of the target frame in a left-to-right and top-to-bottom order; Taking any unmarked disparity point as the origin, determining whether the disparity value of the disparity point within the search radius is within the disparity distribution interval, If yes, mark the disparity point as a first disparity point, if no, mark the disparity point as a second disparity point; The disparity growth frame is determined based on a minimum circumscribed rectangle of the first disparity point.

10. An electronic device, characterized in that: include: A memory storing execution instructions; as well as A processor, wherein the processor executes the execution instructions stored in the memory, so that the processor executes the parallax-based ground reflection misrecognition detection method according to any one of claims 1 to 5.

Images