Shielding detection method for vehicle-mounted electronic outside rear-view mirror

Abstract

The invention relates to the technical field of intelligent auxiliary driving, and provides a vehicle-mounted electronic outside rear-view mirror shielding detection method, based on the vehicle-mounted electronic outside rear-view mirror shielding detection requirement, on one hand, an acquired picture is simplified into a single-channel target image through image processing, image compression is performed, interference frame filtering is executed, and the image is compressed; therefore, while image features (main structures and edge information of the images can still be reserved) are reserved, the data volume is greatly reduced, the detection and recognition efficiency is further improved, and the cost is reduced; on the other hand, texture feature extraction processing is performed to obtain a texture image, a key texture mode is reserved, and irrelevant details are filtered to improve the subsequent processing efficiency; according to the method, the image processing technology is optimized, texture restoration is carried out in combination with texture expansion processing, the model robustness is enhanced by analyzing surrounding texture features and naturally expanding missing parts, so that false alarm and missing alarm of abnormal shielding are reduced, and dependence on hardware resources and cost investment are reduced by optimizing the image processing technology.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent auxiliary driving, and in particular to a vehicle-mounted electronic outside rearview mirror blocking detection method. BACKGROUND

[0002] With the development of intelligent networked automobile technology, electronic outside rearview mirrors gradually replace traditional physical rearview mirrors and become an industry trend. Electronic outside rearview mirrors usually use cameras to collect images and present them through a display screen, but in actual application, the camera of the electronic outside rearview mirror is prone to mounting blocking or blocking, resulting in poor picture display effect and poor driver experience.

[0003] In the prior art, lens blocking is mainly identified in two ways:

[0004] (1) A deep learning algorithm is used, which has high recognition accuracy but also has high requirements for hardware resources, and requires a large number of pre-labeled samples, resulting in high investment costs. It has poor adaptability and insufficient generalization ability for scenes that have not been collected.

[0005] Specifically, since the deep learning algorithm requires a large amount of labeled data to train the model, but the types of blocking in actual road scenes are diverse (such as rain, insect carcasses, and dust), the labeling cost is high and the coverage is not complete, resulting in a sharp drop in recognition accuracy of the model on unobserved blocking types. At the same time, the deep learning method usually has high computational complexity, and the vehicle-mounted edge device has limited computing power, making it difficult to meet the real-time processing requirements, resulting in blocking recognition delay and affecting driving safety.

[0006] (2) An image recognition algorithm is used to detect the blocking area. The image recognition algorithm relies on manually designed features and has consistent recognition results for various scenes, and has low requirements for hardware resources, but the overall recognition accuracy is lower than that of the deep learning method optimized for scenes.

[0007] Specifically, the image algorithm relies on specific scene assumptions and has poor adaptability to complex lighting and dynamic blocking. For example, in a backlight or low-contrast environment, the algorithm is prone to misjudging normal images as blocking. SUMMARY

[0008] The present application provides a vehicle-mounted electronic outside rearview mirror blocking detection method, which solves the technical problems of the two existing blocking detection and recognition schemes based on traditional image algorithms and convolutional neural networks. The former has lower accuracy than the latter, and requires a large number of template matching and rule matching for post-processing. The latter has high accuracy in specific scenes, but is limited by the large number of parameters and computational complexity of the convolutional neural network, resulting in high hardware costs, high data collection and labeling investment costs, and low recognition accuracy and sensitivity due to weak lens blocking recognition robustness and high false positive rate.

[0009] To solve the above technical problems, the application provides a vehicle-mounted electronic outside rearview mirror shielding detection method, comprising the steps of:

[0010] acquiring a collection picture of a vehicle-mounted electronic outside rearview mirror, performing image processing to obtain a single-channel target image;

[0011] performing image processing on the target image, performing image compression and executing interference frame filtering based on image pixels to obtain a to-be-detected image;

[0012] acquiring the filtered to-be-detected image, performing texture feature extraction processing to obtain a texture image, and performing texture repair in combination with texture expansion processing;

[0013] calculating a shielding area ratio according to the repaired texture image, and then judging whether there is abnormal shielding according to a frame difference algorithm.

[0014] The basic scheme is based on the shielding detection requirement of a vehicle-mounted electronic outside rearview mirror. On the one hand, the collection picture is simplified into a single-channel target image through image processing, and image compression and interference frame filtering are performed, so that the data quantity is greatly reduced while the image features (the main structure and edge information of the image can be retained) are retained, and the detection and recognition efficiency is improved and the cost is reduced. On the other hand, texture feature extraction processing is performed to obtain a texture image, key texture patterns are retained, irrelevant details are filtered to improve the subsequent processing efficiency, and texture repair is performed in combination with texture expansion processing. The missing part is naturally expanded by analyzing the surrounding texture features, so as to enhance the model robustness, reduce the false alarm and the missed alarm of abnormal shielding, and reduce the dependence on hardware resources and the cost input through optimization of image processing technology.

[0015] In a further embodiment, the shielding area ratio is calculated according to the repaired texture image, and then it is judged whether there is abnormal shielding according to a frame difference algorithm, comprising:

[0016] A1, acquiring the repaired texture image, and performing grid division to obtain a plurality of grid regions;

[0017] A2, identifying whether each grid region is shielded based on pixel value summation filtering, and counting the area ratio of the shielded grid region relative to the texture image;

[0018] judging whether shielding occurs in the current frame according to the area ratio, and if so, performing frame difference statistics based on the image position of the shielding region to judge whether there is abnormal shielding, and outputting a first conclusion;

[0019] A3, calculating the LOF value of each grid region to assist in judging whether it is shielded, and counting the area ratio of the shielded grid region relative to the texture image;

[0020] According to the area ratio, it is judged whether the current frame appears an occlusion situation, if yes, frame difference statistics is performed based on the image position of the occlusion region, it is judged whether there is an abnormal occlusion, and a second conclusion is output.

[0021] A4、If the first conclusion and the second conclusion are consistent, it is judged that there is an abnormal occlusion.

[0022] In the abnormal occlusion detection and judgment, the pixel value summation filtering recognition is used as a lightweight detection, the pixel statistical difference (such as the occlusion may cause the overall darkening or color mutation) between the normal state and the occluded state is compared, the threshold is set to confirm the occlusion region, the calculation amount is small, and the real-time performance is strong. Meanwhile, the LOF value auxiliary judgment is supplemented, the occlusion is recognized by analyzing the density deviation of the local region of the image, the data distribution is automatically adapted, the unknown occlusion type can be detected, and the rain, fog, backlight, cold region and other harsh scenes can be adapted. Finally, the pixel value summation filtering recognition and the LOF value auxiliary judgment are introduced to perform frame difference statistics and comparison verification, avoiding the interference items such as shooting abnormalities, improving the calculation efficiency while ensuring the detection accuracy.

[0023] In a further embodiment, whether each of the grid regions is occluded is judged based on the pixel value summation filtering recognition, and the area ratio of the occluded grid regions relative to the texture image is counted. Specifically, the pixel value sum of each of the grid regions is counted, it is judged whether the pixel value sum is greater than a preset sum threshold, if yes, the corresponding grid region is marked as 1, indicating that it is occluded, otherwise, it is marked as 0. The total number of the marked grid regions is counted, and the area ratio is calculated in combination with the total number of the regions of the texture image.

[0024] The suspicious region is quickly screened by the pixel summation, the real-time performance is high, and the cost is low.

[0025] In a further embodiment, whether each of the grid regions is occluded is judged based on the LOF value auxiliary judgment, and the area ratio of the occluded grid regions relative to the texture image is counted. Specifically, the LOF value of each of the grid regions is calculated, it is judged whether the corresponding grid region is inconsistent with the density of the surrounding grid regions according to the LOF value, if yes, the corresponding grid region is marked as 1, indicating that it is occluded, otherwise, it is marked as 0. The total number of the marked grid regions is counted, and the area ratio is calculated in combination with the total number of the regions of the texture image.

[0026] In the present scheme, the LOF value is used to judge whether the corresponding grid region is inconsistent with the density of the surrounding grid regions, and whether it is occluded, which can adapt to complex occlusion situations and has good robustness.

[0027] In a further embodiment, whether the current frame appears to be occluded is determined according to the area ratio, and if so, frame difference statistics are performed based on the image position of the occluded area to determine whether there is abnormal occlusion, specifically:

[0028] According to the area ratio and the preset occlusion threshold, whether the current frame appears to be occluded is determined, and when it is determined that the first conclusion / second conclusion determines that there is abnormal occlusion, it is further determined whether the texture images of consecutive multiple frames all appear to be occluded, and if so, the image position of the occluded area on each frame of the texture image is obtained, and the image position is subjected to frame difference statistics, and if the positions are consistent, it is determined that there is abnormal occlusion.

[0029] The present scheme is based on multi-feature input of pixel value summation filtering identification and LOF value auxiliary judgment, and further calculates the area ratio of the occluded area to the total area of the rearview mirror through grid segmentation, and if the area ratio exceeds the preset occlusion threshold, it is determined that occlusion occurs. By reducing the data dimension, the calculation efficiency is high. Based on the persistence of actual occlusion, whether the texture images of consecutive multiple frames all appear to be occluded and whether the image positions of the occluded areas on the texture images are consistent are determined to avoid false positives, thereby improving the anti-interference ability and reliability of occlusion detection.

[0030] In a further embodiment, the repaired texture image is obtained, and a plurality of grid regions are obtained by grid division, specifically: the repaired texture image is obtained, the texture image is subjected to normalization processing, and the texture image is subjected to grid division according to a preset division ratio to obtain a plurality of grid regions.

[0031] The present scheme eliminates the influence of uneven illumination on texture features by normalizing the texture image, reducing environmental interference. Dividing the texture image into grid regions and focusing on the local significantly improves the accuracy and adaptability of occlusion detection, especially suitable for partial occlusion scenes in complex environments, thereby improving the adaptability of occlusion detection to different environments.

[0032] In a further embodiment, the acquisition picture of the vehicle-mounted electronic outside rearview mirror is obtained, and the target image of a single channel is obtained by image processing, including:

[0033] The acquisition picture of the vehicle-mounted left / right electronic outside rearview mirror is obtained;

[0034] Decoding is performed to de-serialize the acquisition picture into an image layer in YUV420 data format;

[0035] The pixel size format of the image layer is converted, and the single-channel image of the Y component of the image is stored as a target image in an image queue.

[0036] The scheme realizes pixel compression through two steps of picture de-serialization into YUV420 data format image layer and pixel size format conversion, while retaining image features, greatly reducing data volume, and further reducing hardware cost and improving computing efficiency.

[0037] In further embodiments, the target image is subjected to image processing, image compression based on image pixels, and interference frame filtering to obtain a to-be-detected image, including:

[0038] The target image is obtained from the image queue, and a right shift operation by 4 bits is performed on the image pixel value of the target image to realize image compression.

[0039] Each pixel point in the compressed target image is traversed, and the image pixel value is counted to determine the pixel value with the most pixel points as the reference value.

[0040] It is judged whether the reference value is greater than the pixel filtering threshold value, if yes, the compressed target image is output as the to-be-detected image, if not, the target image is filtered.

[0041] The scheme realizes image compression by performing a right shift operation by 4 bits on the image pixel value, directly reducing the storage space by 50%, and has good real-time performance and extremely low resource consumption.

[0042] In further embodiments, texture feature extraction processing is performed to obtain a texture image, including:

[0043] A horizontal direction integral operation is performed on the to-be-detected image to extract texture features and obtain a horizontal gradient component to obtain a horizontal feature image.

[0044] A vertical direction integral operation is performed on the to-be-detected image to extract texture features and obtain a vertical gradient component to obtain a vertical feature image.

[0045] The intensity of edge texture is calculated according to the horizontal feature image and the vertical feature image to obtain a texture image.

[0046] The scheme simultaneously performs horizontal direction integral operation and vertical direction integral operation on the to-be-detected image to extract texture features, calculates the intensity of edge texture according to the horizontal feature image and the vertical feature image to obtain a texture image, and has high computing efficiency through joint directional analysis and statistical features. While ensuring speed, structured information of texture can be effectively captured (through texture extraction, the integral curve of the obscured area of the rearview mirror is significantly different from the clean mirror, which helps subsequent abnormal occlusion recognition).

[0047] In further embodiments, texture repair is performed in combination with texture expansion processing, including:

[0048] The texture image is threshold binarization processed to obtain a gray image;

[0049] The gray image is subjected to an open operation, the image is eroded first to remove noise points in the gray image, and then inflation is performed to enlarge the binarized texture to realize texture repair.

[0050] The threshold binarization processing is performed first to reduce the calculation complexity and facilitate subsequent analysis, the gray image is subjected to an open operation, the image is eroded first to remove noise points in the gray image, and then inflation is performed to enlarge the binarized texture to realize texture repair, the texture of the occluded part is naturally generated by analyzing the surrounding texture, the missing area is filled, and the general outline of the target is maintained. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 is a working flowchart of a vehicle-mounted electronic outside rearview mirror occlusion detection method provided by the embodiment of the present application;

[0052] Figure 2 is a comparison diagram of different scene images provided by the embodiment of the present application. DETAILED DESCRIPTION

[0053] The embodiments of the present application will be specifically described below with reference to the accompanying drawings, and the embodiments are given only for illustrative purposes, and cannot be understood as limiting the present application, including the accompanying drawings for reference and illustration only, and do not constitute a limitation on the protection scope of the present application, because many changes can be made to the present application without departing from the spirit and scope of the present application.

[0054] The vehicle-mounted electronic outside rearview mirror occlusion detection method provided by the embodiment of the present application, as shown in Figure 1 、 Figure 2 , in the embodiment, includes steps S1-S5:

[0055] S1, acquiring a collection picture of a vehicle-mounted electronic outside rearview mirror, performing image processing to obtain a single-channel target image, including:

[0056] S11, acquiring a collection picture of a left / right electronic outside rearview mirror;

[0057] S12, performing decoding to de-serialize the collection picture into an image layer in YUV420 data format;

[0058] Specifically, the collection picture collected by the electronic outside rearview mirror is transmitted through a data stream transmission module of an RTOS system, and is de-serialized into an image layer in YUV420 data format.

[0059] S13, performing pixel size format conversion on the image layer, and storing a single-channel image of the Y component of the acquired image as a target image into an image queue.

[0060] For example, the YUV420 image layer is streamed to the LINUX system through the IPC and the CAN, and the application layer only acquires the 680x360 image buffer to perform format conversion to obtain a single-channel image of the image Y component.

[0061] The embodiment realizes the pixel compression in two steps of image layer conversion into the YUV420 data format and pixel size format conversion, greatly reduces the data volume while retaining the image features, and further reduces the hardware cost and improves the calculation efficiency.

[0062] S2, performing image processing on the target image, performing image compression based on image pixels and executing interference frame filtering to obtain a to-be-detected image, comprising:

[0063] acquiring the target image from the image queue, and performing a right shift operation by 4 bits on the image pixel value of the target image to realize image compression and compress the image pixel value in the range of 0-16;

[0064] traversing each pixel point in the compressed target image, counting the image pixel value, and determining the pixel value with the largest total number of pixel points as a reference value;

[0065] determining whether the reference value is greater than a pixel filtering threshold value, if yes, outputting the compressed target image as the to-be-detected image, and if not, filtering the target image.

[0066] In the embodiment, the calculation formula of image compression is as follows:

[0067]

[0068] In the formula, Y represents the original pixel value of the Y component image acquired from the data queue, and the value range is 0≤Y≤255; n represents the number of bits of right shift, and here n=4, and the purpose is to compress the pixel value to the interval of 0-16; the right shift by n bits is equivalent to dividing the original pixel value by 2 and taking the integer part; f(x) operation represents the number of times of the compressed pixel value appearing in the image. n

[0069] The pixel filtering threshold value can be selected according to the number of bits of right shift, and in the embodiment, the preferred value is 3.

[0070] The embodiment realizes image compression by performing a right shift operation by 4 bits on the image pixel value, and the storage space is directly reduced by 50%, which is good in real-time performance and extremely low in resource consumption.

[0071] S3, acquiring the filtered to-be-detected image, performing texture feature extraction processing to obtain a texture image, and performing texture repair in combination with texture expansion processing;

[0072] ​In the embodiment, texture feature extraction processing is performed to obtain a texture image, including:

[0073] S3a, performing horizontal direction integral operation on the to-be-detected image, performing texture feature extraction to obtain a horizontal gradient component, obtaining a horizontal feature image, and the calculation formula is as follows:

[0074]

[0075] The horizontal direction kernel is as follows:

[0076]

[0077] S3b, performing vertical direction integral operation on the to-be-detected image, performing texture feature extraction to obtain a vertical gradient component, obtaining a vertical feature image, and the calculation formula is as follows:

[0078]

[0079] The vertical direction kernel is as follows:

[0080]

[0081] S3c, calculating the intensity of the edge texture according to the horizontal feature image and the vertical feature image, obtaining a texture image, and the calculation formula is as follows:

[0082]

[0083] In the formula, I(x, y) represents the Y component value of the image at the coordinate (x, y), K x , and K y respectively represent the Sobel convolution kernel in the horizontal and vertical directions, i and j represent the index offset of the convolution kernel (relative to the center pixel, the value range is (-1, 0, 1)), G x (x, y) represents the horizontal gradient component at the image, and G y (x, y) represents the vertical gradient component at the image, and G(x, y) represents the intensity of the edge texture.

[0084] In the embodiment, the horizontal direction integral operation and the vertical direction integral operation are simultaneously performed on the to-be-detected image to perform texture feature extraction, the intensity of the edge texture is calculated according to the horizontal feature image and the vertical feature image, and the texture image is obtained. Through joint directional analysis and statistical features, the calculation efficiency is high, the structured information of the texture can be effectively captured while the speed is ensured (through texture extraction, the integral curve of the blocked area of the rearview mirror is significantly different from the clean mirror, which is helpful for subsequent abnormal blocking identification).

[0085] In the embodiment, texture repair is performed in combination with texture expansion processing, including:

[0086] S3A, threshold binarization processing is performed on the texture image to obtain a grayscale image;

[0087] For example, the pixel value of each pixel point on the texture image is traversed, and if the pixel value is greater than 50, it is set to 255, otherwise it is set to 0.

[0088] S3B, an open operation is performed on the grayscale image, the image is first eroded to remove noise points in the grayscale image, and then dilated to enlarge the binary texture to realize texture repair, and the calculation formula is as follows:

[0089] O = ((I >= 50) * S) O S

[0090] In the formula, I represents an input original grayscale image, and the pixel value range is [0, 255]. *S refers to a texture structure operation, and the above formula represents erosion. O S refers to a texture structure operation, and the above formula represents dilation.

[0091] The embodiment first performs threshold binarization processing to reduce the calculation complexity and facilitate subsequent analysis; the open operation is performed on the grayscale image, the image is first eroded to remove noise points in the grayscale image, and then dilated to enlarge the binary texture to realize texture repair, and the texture of the occluded part is naturally generated by analyzing the surrounding texture, filling the missing area and maintaining the general outline of the target.

[0092] S4, the occlusion area ratio is calculated according to the repaired texture image, and then it is judged whether there is an abnormal occlusion according to the frame difference algorithm, including A1-A4:

[0093] A1, the repaired texture image is obtained, and a plurality of grid regions are obtained by grid division, specifically: the repaired texture image is obtained, the texture image is normalized, and the texture image is divided into a plurality of grid regions according to a preset division ratio.

[0094] For example: the normalized operation is performed on the image, and the value of the binary image is changed from [0, 255] to [0, 1]; the normalized image is divided into grid blocks, the division ratio is 4x3, and the grid region is divided into 12 square regions.

[0095] The embodiment eliminates the influence of uneven light on texture features by normalizing the texture image, and reduces environmental interference; and the texture image is divided into grid regions, which significantly improves the accuracy and adaptability of the occlusion detection, especially suitable for partial occlusion scenes in complex environments, thereby improving the adaptability of the occlusion detection to different environments.

[0096] A2, whether each of the grid area is blocked is identified based on pixel value summation filtering, and the area proportion of the blocked grid area relative to the texture image is counted; whether the current frame appears a blocking condition is judged according to the area proportion, if yes, frame difference statistics is carried out based on the image position of the blocking area, whether there is abnormal blocking is judged, and a first conclusion is output, specifically:

[0097] A21, the pixel value sum of each of the grid areas is counted, whether the pixel value sum is greater than a preset sum threshold sumArea (for example, sumArea> 30) is judged, if yes, the corresponding grid area is marked as 1, indicating that it is blocked, otherwise it is marked as 0; the total number of blocked grid areas is counted, and the area proportion is calculated in combination with the total number of areas of the texture image.

[0098] A22, whether the current frame appears a blocking condition is judged according to the area proportion and a preset blocking threshold, when it is judged that the first conclusion judges that there is abnormal blocking, whether the texture images of continuous multiple frames all appear a blocking condition is further determined, if yes, the image position of the blocking area on each frame of the texture image is obtained, frame difference statistics is carried out on the image position, if the position is consistent, it is judged that there is abnormal blocking.

[0099] In the embodiment, the preset blocking threshold can be set according to actual needs, for example, 50%.

[0100] The embodiment quickly screens suspicious areas through pixel summation, has high real-time performance and low cost.

[0101] A3, the LOF value of each of the grid areas is calculated to assist in judging whether it is blocked, and the area proportion of the blocked grid area relative to the texture image is counted; whether the current frame appears a blocking condition is judged according to the area proportion, if yes, frame difference statistics is carried out based on the image position of the blocking area, whether there is abnormal blocking is judged, and a second conclusion is output, specifically:

[0102] A31, the LOF value of each of the grid areas is calculated, whether the corresponding grid area is inconsistent with the surrounding grid area in density is judged according to the LOF value, if yes, the corresponding grid area is marked as 1, indicating that it is blocked, otherwise it is marked as 0; the total number of blocked grid areas is counted, and the area proportion is calculated in combination with the total number of areas of the texture image.

[0103] LOF is the abbreviation of Local Outlier Factor, which means "local outlier factor", which can calculate the local density deviation of a given data point relative to its neighbors. It considers that the sample with a density much lower than its neighbors is an outlier. The formula of LOF value is as follows:

[0104]

[0105] In the formula:

[0106] p represents the grid block currently requiring calculation of the LOF value;

[0107] q and r represent other grid blocks related to p, which are used to traverse the neighborhood and calculate the distance, density and the like in the calculation process;

[0108] k is a preset integer, which is used to determine the neighborhood range and is taken as 7 in the method;

[0109] d(p, q) represents the Euclidean distance between p and q;

[0110] d(q, r) represents the Euclidean distance between q and r;

[0111] k_distance(r) represents that there is a distance d, such that grid block r has at least k neighbors (including r itself) with a distance from r of not more than d, and at most k-1 neighbors with a distance from r of less than d, which determines the k-neighborhood range of r, and k_distance(q) is the same;

[0112] Nk(p) refers to a set of all grid blocks with a distance from p of less than or equal to k_distance(p), that is, the k-neighborhood of p;

[0113] |Nk(p)| represents the number of grid blocks in the set, that is, the size of the neighborhood, which is used for density calculation and weight allocation;

[0114] Nk(q) is a k-neighborhood determined with a neighbor grid block q of grid block p as the core, which is used to calculate the local reachable density of q;

[0115] |Nk(q)| represents the number of grid blocks in the set, that is, the size of the neighborhood.

[0116] The embodiment adopts the LOF value to judge whether the corresponding grid region is inconsistent with the density of the surrounding grid region, to judge whether it is occluded, which can adapt to complex occlusion conditions and has good robustness.

[0117] A32, according to the area ratio and the preset occlusion threshold, whether the current frame appears an occlusion condition is judged, when the second conclusion is judged to exist an abnormal occlusion, whether the texture images of continuous multiple frames all appear an occlusion condition is further determined, if yes, the image positions of the occlusion regions on each frame of the texture images are acquired, frame difference statistics is performed on the image positions, if the positions are consistent, it is judged that there exists an abnormal occlusion.

[0118] The embodiment is based on pixel value summation filtering identification, LOF value auxiliary judgment and multi-feature input, further calculates the area ratio of the shielding area to the total area of the rearview mirror through grid segmentation, and determines that the shielding condition occurs if the area ratio exceeds the preset shielding threshold. The calculation efficiency is high by reducing the data dimension. Based on the persistence of actual shielding, whether the shielding condition occurs in the continuous multiple frames of the texture image and whether the image positions of the shielding area on the texture image are consistent are judged to avoid triggering false positives, thereby improving the anti-interference ability and reliability of the shielding detection.

[0119] A4, if the first conclusion and the second conclusion are consistent, it is determined that there is an abnormal shielding.

[0120] In the embodiment, steps A2 and A3 only act as an explanation and do not limit the sequence of the steps.

[0121] In the embodiment, when performing abnormal shielding detection and judgment, the pixel value summation filtering identification is used as a lightweight detection, the threshold of the shielding area is set by comparing the pixel statistical difference between the normal state and the shielding state (such as shielding may cause overall darkening or color mutation), the calculation amount is small, and the real-time performance is strong; meanwhile, the LOF value auxiliary judgment is supplemented, the shielding is identified by analyzing the density deviation of the local area of the image, the data distribution is automatically adapted, unknown shielding types can be detected, and adverse scenes such as rain, fog, backlight, cold region and the like can be adapted; finally, the pixel value summation filtering identification and the LOF value auxiliary judgment are introduced to perform frame difference statistics and comparison verification, and interference items such as shooting abnormalities are avoided, the detection accuracy is considered, and the calculation efficiency is improved.

[0122] S5, an alarm prompt is given according to the abnormal shielding, including but not limited to voice reminding, central control screen reminding and sound and light alarm.

[0123] All experiments take scene detection rate and scene false detection rate as evaluation indexes, test cold region, rainy day, haze, parking lot, normal road, dark light and the like, wherein, 50 normal scenes and 50 artificial shielding scenes are included, and a total of 100 test scenes are included.

[0124] After the test, the following results are obtained:

[0125] Method Detection example Error detection example Detection rate Error detection rate CoverDet 44 7 88％ 14％ Ours 43 4 86％ 8％

[0126] In the above table, the detection instance is 50 artificial shielding scenes correctly detected, the detection rate is the proportion of 50 artificial shielding scenes correctly detected, the false detection instance is 50 normal scenes wrongly detected, the false detection rate is the proportion of 50 normal scenes wrongly detected, the false detection rate is lower, and the priority is higher than the detection rate.

[0127] CoverDet is a detection method in the deep learning mode, and from the above table, it can be seen that the detection rate of the deep learning mode is higher than that of the method proposed in the embodiment, but the two are close; the detection rate of the deep learning method is limited by the prior knowledge of data training, and there is a higher false detection for the untrained scene, and the false detection rate of the method proposed in the embodiment is much lower than that of the deep learning method, and the robustness is higher.

[0128] The random field scene test is carried out using the scheme, and the effect is as Figure 2 , from left to right in the figure are cold region scene, parking lot scene, parking lot scene, artificial occlusion scene, from top to bottom are the collected picture, gray image and grid region marking picture.

[0129] The area ratio of the cold region scene is 5 / 12, which is less than 50%, so it is judged as no occlusion, and the detection and recognition are the same as the actual scene.

[0130] The area ratio of the parking lot scene is 4 / 12, which is less than 50%, so it is judged as no occlusion, and the detection and recognition are the same as the actual scene.

[0131] The area ratio of the artificial occlusion scene is 8 / 12, which is greater than 50%, so it is judged as abnormal occlusion, and the detection and recognition are the same as the actual scene.

[0132] Based on the test conclusions of the above three scenes, the occlusion detection of the embodiment has high robustness, good adaptability, and high detection accuracy.

[0133] The embodiment of the application is based on the occlusion detection requirement of the vehicle-mounted electronic outer rearview mirror, on the one hand, the collected picture is simplified into a single-channel target image through image processing, and image compression and interference frame filtering are performed, so that the data amount is greatly reduced while the image features (the main structure and edge information of the image can be retained) are retained, thereby improving the detection and recognition efficiency and reducing the cost; on the other hand, texture feature extraction processing is performed to obtain a texture image, key texture patterns are retained, irrelevant details are filtered to improve the subsequent processing efficiency; and combined with texture expansion processing, texture repair is performed, the missing part is naturally expanded by analyzing the surrounding texture features, so as to enhance the model robustness, reduce the false alarm and the false negative of abnormal occlusion, and reduce the dependence on hardware resources and the cost input through optimization of image processing technology.

[0134] The above embodiment is a preferred embodiment of the application, but the embodiment of the application is not limited by the above embodiment, any change, modification, substitution, combination, simplification made without departing from the spirit and principle of the application should be an equivalent replacement mode, and all are included in the protection scope of the application.

Claims

1. A method of detecting a blocking of an electronic exterior mirror of a vehicle, characterized in that The method comprises the steps of: acquiring a collection picture of an electronic outside rearview mirror of a vehicle, and performing image processing to obtain a single-channel target image; performing image processing on the target image, performing image compression based on image pixels, and performing interference frame filtering to obtain a to-be-detected image; acquiring the filtered to-be-detected image, performing texture feature extraction processing to obtain a texture image, and performing texture repair in combination with texture expansion processing; calculating an occlusion area ratio based on the repaired texture image, and then judging whether there is abnormal occlusion based on a frame difference calculation method.

2. The method of claim 1, wherein the method further comprises: According to the repaired texture image, the occlusion area ratio is calculated, and then it is judged whether there is abnormal occlusion according to the frame difference calculation method, comprising: A1, acquiring the repaired texture image, and performing grid division to obtain a plurality of grid regions; A2, based on pixel value summation filtering, identifying whether each grid region is occluded, and counting the area ratio of the occluded grid regions relative to the texture image; judging whether the current frame appears an occlusion situation according to the area ratio, and if so, performing frame difference statistics based on the image position of the occluded region to judge whether there is abnormal occlusion, and outputting a first conclusion; A3, calculating the LOF value of each grid region to assist in judging whether it is occluded, and counting the area ratio of the occluded grid regions relative to the texture image; judging whether the current frame appears an occlusion situation according to the area ratio, and if so, performing frame difference statistics based on the image position of the occluded region to judge whether there is abnormal occlusion, and outputting a second conclusion; A4, if the first conclusion and the second conclusion are consistent, it is judged that there is abnormal occlusion.

3. The method of claim 2, wherein the method further comprises: Based on pixel value summation filtering, whether each grid region is occluded is identified, and the area ratio of the occluded grid regions relative to the texture image is counted. Specifically, the pixel value sum of each grid region is counted, it is judged whether the pixel value sum is greater than a preset sum threshold, if so, the corresponding grid region is marked as 1, indicating that it is occluded, otherwise it is marked as 0; the total number of marked grid regions is counted, and the area ratio is calculated in combination with the total number of regions of the texture image.

4. The method of claim 2, wherein the method further comprises: The LOF value of each grid region is calculated to assist in judging whether it is occluded, and the area ratio of the occluded grid regions relative to the texture image is counted. Specifically, the LOF value of each grid region is calculated, it is judged whether the corresponding grid region is inconsistent with the surrounding grid regions in density according to the LOF value, if so, the corresponding grid region is marked as 1, indicating that it is occluded, otherwise it is marked as 0; the total number of marked grid regions is counted, and the area ratio is calculated in combination with the total number of regions of the texture image.

5. The method of claim 2, wherein the method further comprises: According to the area ratio, it is judged whether the current frame appears an occlusion situation, and if so, frame difference statistics is performed based on the image position of the occluded region to judge whether there is abnormal occlusion. According to the area ratio and the preset occlusion threshold, it is judged whether the current frame appears an occlusion situation, when it is judged that the first conclusion / second conclusion judges that there is an abnormal occlusion, it is further determined whether the texture image of the continuous multiple frames appears an occlusion situation, if yes, the image position of the occlusion area on each frame of the texture image is obtained, and the frame difference statistics is performed on the image position, if the position is consistent, it is judged that there is an abnormal occlusion.

6. The method of claim 2, wherein the method further comprises: The repaired texture image is obtained, and grid division is performed to obtain a plurality of grid regions, specifically: the repaired texture image is obtained, the texture image is normalized, and the texture image is divided into a plurality of grid regions according to a preset division ratio.

7. The method of claim 1, wherein the method further comprises: The acquisition picture of the vehicle-mounted electronic outside rearview mirror is obtained, and the target image of a single channel is obtained by image processing, including: The acquisition picture of the vehicle-mounted left / right electronic outside rearview mirror is obtained; Decoding is performed to unstring the acquisition picture into an image layer in YUV420 data format; The pixel size format of the image layer is converted, and the single-channel image of the image Y component is stored as the target image in the image queue.

8. The method of claim 1, wherein the method further comprises: The target image is processed, the image is compressed based on the image pixel, and interference frame filtering is performed to obtain a to-be-detected image, including: The target image is obtained from the image queue, and the image pixel value of the target image is right-shifted by 4 bits to realize image compression; Each pixel point in the compressed target image is traversed, and the image pixel value is counted to determine the pixel value with the largest total number of pixel points as a reference value; It is judged whether the reference value is greater than a pixel filtering threshold, if yes, the compressed target image is output as the to-be-detected image, if not, the target image is filtered.

9. The method of claim 8, wherein the method further comprises: Texture feature extraction processing is performed to obtain a texture image, including: The to-be-detected image is subjected to horizontal direction integration operation, texture feature extraction is performed to obtain horizontal gradient components, and a horizontal feature image is obtained; The to-be-detected image is subjected to vertical direction integration operation, texture feature extraction is performed to obtain vertical gradient components, and a vertical feature image is obtained; The intensity of the edge texture is calculated according to the horizontal feature image and the vertical feature image, and a texture image is obtained.

10. The method of claim 1, wherein the method further comprises: determining whether the electronic mirror is in a folded position; and determining whether the electronic mirror is in a stowed position. 10 Texture repair is performed in combination with texture expansion processing, including: The texture image is subjected to threshold binaryzation processing to obtain a gray-scale image; The gray-scale image is subjected to opening operation, the image is first eroded to remove noise points in the gray-scale image, and then dilatation is performed to enlarge the binaryzation texture to realize texture repair.

Images