Blind sidewalk occupation detection method, device and equipment and storage medium

Abstract

The invention provides a blind sidewalk occupation detection method and device, equipment and a storage medium, and the method comprises the steps: obtaining a to-be-detected target image, and recognizing a blind sidewalk region in the target image; a breakpoint area located in the blind sidewalk area is detected in the target image, a plurality of area sub-images are intercepted from the target image on the basis of the breakpoint area, and intersection exists between each area sub-image and the breakpoint area; and identifying blind sidewalk occupation prediction results represented by the area sub-images, and judging whether the blind sidewalk area in the target image is occupied according to the blind sidewalk occupation prediction results. According to the technical scheme provided by the invention, the blind sidewalk occupation detection efficiency can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technology, and in particular to a blind path occupancy detection method, device, equipment and storage medium. Background Art

[0002] There are currently a variety of methods for detecting whether a blind path is occupied. For example, a camera can be deployed near the blind path to shoot the blind path in real time. Subsequently, a human can identify whether the blind path is occupied in the captured video. This method is not intelligent enough because it requires real-time human detection.

[0003] For another example, an object recognition model can be pre-trained, and then the image information of the blind path can be collected and processed by the object recognition model to identify whether the blind path is occupied by bicycles, electric vehicles or other obstacles. This method can automatically detect whether the blind path is occupied, but due to the large number of possible obstacles on the blind path, the object recognition model cannot effectively determine whether the blind path is occupied for obstacles that are not involved in the model training stage.

[0004] In view of this, a more efficient blind path occupancy detection method is currently needed. Summary of the invention

[0005] In view of this, one or more embodiments of the present disclosure provide a blind path occupancy detection method, apparatus, device and storage medium, which can improve the efficiency of blind path occupancy detection.

[0006] On one hand, the present disclosure provides a method for detecting blind path occupancy, the method comprising: acquiring a target image to be detected, and identifying a blind path area in the target image; detecting a breakpoint area located in the blind path area in the target image, and based on the breakpoint area, intercepting a plurality of area sub-graphs from the target image, wherein each of the area sub-graphs has an intersection with the breakpoint area; identifying a blind path occupancy prediction result represented by each of the area sub-graphs, and judging whether the blind path area in the target image is occupied according to the blind path occupancy prediction result.

[0007] On the other hand, the present disclosure further provides a device for detecting blind path occupancy, the device comprising: a region identification unit, used to acquire a target image to be detected and identify a blind path region in the target image; a sub-image capture unit, used to detect a breakpoint region located in the blind path region in the target image, and based on the breakpoint region, capture a plurality of region sub-images from the target image, wherein each of the region sub-images has an intersection with the breakpoint region; and a judgment unit, used to identify a blind path occupancy prediction result represented by each of the region sub-images, and judge whether the blind path region in the target image is occupied according to the blind path occupancy prediction result.

[0008] Another aspect of the present disclosure provides an electronic device, which includes a memory and a processor, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the above-mentioned blind path occupancy detection method is implemented.

[0009] On the other hand, the present disclosure further provides a computer-readable storage medium, which is used to store a computer program. When the computer program is executed by a processor, the above-mentioned blind path occupancy detection method is implemented.

[0010] The technical solution provided by one or more embodiments of the present disclosure does not need to pay attention to the type of obstacles occupying the blind path, but only detects the breakpoint area in the blind path area, and based on the breakpoint area, screenshots of multiple different area sub-graphs are taken. Subsequently, by analyzing the blind path occupancy prediction results represented by each area sub-graph, it can be comprehensively determined whether the current blind path area is occupied.

[0011] The above-mentioned breakpoint area can reflect the discontinuity of the blind path area. When the blind path area in the target image shows discontinuity characteristics, it indicates that there may be obstacles in the blind path area. By analyzing the breakpoint area, the actual type of obstacles in the blind path area can be ignored, making the technical solution of the present disclosure more universal in blind path occupancy detection. At the same time, the above process can automatically process the target image and output the final judgment result, thereby improving the efficiency of blind path occupancy detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The features and advantages of the embodiments of the present disclosure may be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present disclosure in any way. In the accompanying drawings:

[0013] Figure 1 A first example schematic diagram of blind path occupancy is shown;

[0014] Figure 2 A schematic diagram of the steps of blind path occupancy detection in one embodiment of the present disclosure is shown;

[0015] Figure 3 A schematic diagram of a process for detecting blind path occupancy in one embodiment of the present disclosure is shown;

[0016] Figure 4 A schematic diagram showing a breakpoint area in one embodiment of the present disclosure is shown;

[0017] Figure 5 A second example schematic diagram of blind path occupancy is shown;

[0018] Figure 6 A schematic diagram of a region sub-graph in one embodiment of the present disclosure is shown;

[0019] Figure 7 A schematic diagram of functional modules of a blind path occupancy detection device in one embodiment of the present disclosure is shown;

[0020] Figure 8 A schematic structural diagram of an electronic device in one embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0022] In the related art, the image information of the blind path can be processed by an object recognition model to identify whether there are bicycles, electric vehicles or other obstacles on the blind path. The limitation of this method is that the object recognition model can usually only recognize the types of obstacles involved in the training phase. For example, in the training phase of the object recognition model, by providing bicycle samples, electric vehicle samples or other obstacle samples, the trained object recognition model can more accurately identify these types of obstacles. However, for obstacle types that are not involved in the training phase, such as trash cans, packages, etc., the object recognition model cannot accurately identify these obstacles in the blind path. It can be seen that this blind path occupancy detection method in the related art depends on the sample type and number of samples provided in the training phase, and has poor scalability.

[0023] The blind path occupancy detection method provided in an embodiment of the present disclosure may not judge the obstacle type, but analyze based on the general features of the blind path when it is occupied. Specifically, when the blind path is not occupied, it is usually a continuous area in the captured image. However, if the blind path is occupied, the feature reflected in the image is that the blind path is separated by an object, and the visual effect is that the blind path is not continuous. For example, in Figure 1 In the example, the blind path is occupied by a trash can, so from a visual effect, the blind path is separated by the trash can. In view of this, the breakpoint area in the blind path area can be analyzed to determine whether the blind path area is occupied.

[0024] Specifically, see Figure 2 and Figure 3 , a blind path occupancy detection method provided by an embodiment of the present disclosure may include the following steps.

[0025] S1: Acquire a target image to be detected, and identify a blind path area in the target image.

[0026] In this embodiment, the target image may be an image including a blind path area. By analyzing the target image, it can be determined whether the blind path area in the target image is occupied. In practical applications, there are many ways to collect the target image. For example, it can be obtained by shooting with a camera deployed near the blind path, or by shooting with a drone with image acquisition function, or by a user and uploading it to the blind path occupancy detection platform. The present disclosure does not limit the method of collecting the target image.

[0027] In this embodiment, the blind road segmentation model may be used to process the target image, thereby obtaining a blind road region recognition result of the target image. The blind road region recognition result may include one or more recognized blind road regions.

[0028] In practical applications, the blind path segmentation model can have multiple implementation forms. For example, the blind path segmentation model can be implemented through network architectures such as HRNet, maskrcnn, solov2, and deepmask. In practical applications, the corresponding network architecture can be flexibly selected according to different scene requirements.

[0029] During the training of the blind road segmentation model, a blind road image sample may be obtained first. The blind road image sample has a binary image label. The binary image label distinguishes the blind road area and the non-blind road area in the blind road image sample by different pixel values.

[0030] Specifically, the blind path image sample may be an image containing a blind path area obtained by taking an image acquisition device. For the blind path image sample, the blind path area may be labeled, wherein the blind path area may be characterized by a first pixel value, and the non-blind path area may be characterized by a second pixel value, thereby obtaining a binary image label of the blind path image sample. The binary image label may have the same size and resolution as the blind path image sample, and through the binary image label, the blind path area and the non-blind path area in the blind path image sample may be distinguished by the two different pixel values ​​of the first pixel value and the second pixel value.

[0031] In this embodiment, the blind road image sample is processed by the blind road segmentation model to generate a predicted binary image corresponding to the blind road image sample. In the predicted binary image, the blind road area can also be represented by the above-mentioned first pixel value, and the non-blind road area can be represented by the above-mentioned second pixel value. During the training process, the predicted binary image output by the blind road segmentation model may not be accurate, so there will be a certain error between the predicted binary image and the binary image label.

[0032] In this embodiment, the error information between the predicted binary image and the binary image label can be generated by the MSE (Mean-Squared Error) Loss function or other error functions. Subsequently, the blind road segmentation model can be corrected based on the error information, so that after training with a large number of blind road image samples, the predicted binary image output by the blind road segmentation model can tend to be consistent with the binary image label.

[0033] In this embodiment, by processing the target image, a blind path area recognition result including one or more blind path areas can be obtained. It should be noted that in some scenarios, some areas in the target image may be misjudged as blind path areas. In view of this, it is necessary to further screen the various blind path areas in the blind path area recognition result to obtain accurate blind path areas.

[0034] Specifically, the actual blind path area usually accounts for a high proportion in the target image. In view of this, for any target blind path area in the blind path area recognition result, the area proportion of the target blind path area in the target image can be determined. If the area proportion is less than or equal to the specified ratio threshold, it indicates that the target blind path area is a misjudgment, and the target blind path area can be eliminated from the blind path area recognition result. If the area proportion is greater than the specified ratio threshold, it indicates that the target blind path area is a real blind path area, and the target blind path area can be retained in the blind path area recognition result.

[0035] For each blind path area in the blind path area recognition result, the above-mentioned area proportion judgment can be performed, so as to retain the real blind path area in the blind path area recognition result.

[0036] Of course, in practical applications, the blind path area in the blind path area recognition result can also be screened by combining various information. For example, the blind path area can be further determined by combining the color of the blind path area and / or the shape of the bricks in the blind path area.

[0037] S3: Detecting a breakpoint region in the blind path region in the target image, and based on the breakpoint region, extracting a plurality of region sub-images from the target image, wherein each of the region sub-images has an intersection with the breakpoint region.

[0038] Considering that if the blind path is occupied, the feature reflected in the image is that the blind path is separated by an object, and the visual effect is that the blind path is not continuous. This discontinuous feature will be reflected in the image in the form of breakpoints. When the blind path is occupied, in the target image, the outline of the obstacle and the blind path area will form a dividing line, and the pixels on the dividing line can be used as the above breakpoints. For example, in Figure 1 In the example, when the blind path is occupied by a trash can, the outline of the trash can and the blind path area will form a dividing line, and the pixel points on the dividing line can be used as breakpoints in the blind path area.

[0039] In this embodiment, the target image can be processed using the breakpoint detection model to obtain the breakpoint prediction result corresponding to the target image. Similar to the training process of the blind path segmentation model in step S1, the training process of the breakpoint detection model can also be implemented with the help of binary image labels.

[0040] Specifically, a breakpoint image sample may be obtained first, wherein the breakpoint image sample has a binary image label, and the binary image label distinguishes the breakpoint area and the non-breakpoint area in the breakpoint image sample by different pixel values.

[0041] Specifically, the breakpoint image sample can be any image containing a breakpoint area. It should be noted that, during the training phase, the breakpoint image sample does not only represent the breakpoints in the blind path area, but can represent the breakpoints in any area. For example, there are lane lines on the road. When the lane lines are blocked by moving vehicles, a breakpoint area of ​​the lane lines will also be formed. Such images can also be used as breakpoint image samples. In other words, in order to improve the generalization ability of the breakpoint detection model, during the training phase, the breakpoint image samples are not limited to representing the breakpoints in the blind path area.

[0042] In this embodiment, for the obtained breakpoint image sample, the breakpoint area can be labeled, wherein each pixel in the breakpoint area can be characterized by a first pixel value, and each pixel in the non-breakpoint area can be characterized by a second pixel value, thereby obtaining a binary image label of the breakpoint image sample. The binary image label can have the same size and resolution as the breakpoint image sample, and through the binary image label, the breakpoint area and the non-breakpoint area in the breakpoint image sample can be distinguished by the two different pixel values ​​of the first pixel value and the second pixel value.

[0043] In this embodiment, the breakpoint image sample is processed by the breakpoint detection model to generate a predicted binary image corresponding to the breakpoint image sample. In the predicted binary image, the breakpoint area can also be represented by the above-mentioned first pixel value, and the non-breakpoint area can be represented by the above-mentioned second pixel value. During the training process, the predicted binary image output by the breakpoint detection model may not be accurate, so there will be a certain error between the predicted binary image and the binary image label.

[0044] In this embodiment, the error information between the predicted binary image and the binary image label can be generated by the MSE Loss function or other error functions. Subsequently, the breakpoint detection model can be corrected based on the error information, so that after training with a large number of breakpoint image samples, the predicted binary image output by the breakpoint detection model can tend to be consistent with the binary image label of the breakpoint image sample.

[0045] In this embodiment, the target image is processed using the trained breakpoint detection model to obtain a breakpoint prediction result corresponding to the target image, and the breakpoint prediction result can characterize the position of the candidate breakpoint in the target image. Since the candidate breakpoints predicted by the breakpoint detection model are not necessarily breakpoints in the blind path area, it is necessary to further process the positions of the candidate breakpoints to screen out the candidate breakpoints in the blind path area.

[0046] Specifically, the candidate breakpoints outside the blind path area may be removed from the breakpoint prediction result, and the area formed by the candidate breakpoints retained in the breakpoint prediction result is used as the detected breakpoint area within the blind path area.

[0047] In practical applications, the breakpoint area formed by the breakpoints can be a rectangular area that just covers the breakpoints. Figure 4 In the figure, each breakpoint forms a curve, then the rectangular area that can just cover the curve (as shown by the dotted line) can be used as the breakpoint area formed by these breakpoints.

[0048] In this embodiment, after the breakpoint area located in the blind path area is detected from the target image, it is necessary to further confirm whether the breakpoint area is referenced by an obstacle occupying the blind path. Figure 1 In the example, the breakpoint area is caused by the trash can occupying the blind path, but in Figure 5 In the example, the breakpoint area between the advertising column and the blind path area is actually caused by the shooting angle of the target image. In other words, detecting the breakpoint area in the blind path area from the target image cannot directly determine whether the blind path area is occupied by obstacles.

[0049] In this embodiment, in order to further determine whether the blind path area is occupied based on the breakpoint area, multiple area sub-images that intersect with the breakpoint area can be intercepted from the target image based on the breakpoint area, and then it can be determined whether each area sub-image represents that the blind path is occupied, and based on the final summary result, it is comprehensively determined whether the blind path area in the target image is occupied.

[0050] In this embodiment, the multiple area sub-images intercepted from the target image can have different sizes, and these area sub-images can reflect the local characteristics of the breakpoint area and the overall environmental characteristics near the breakpoint area. By judging these area sub-images separately, the final result of whether the blind path area is occupied can be made more accurate.

[0051] In practical applications, when capturing multiple regional sub-images from a target image, there may be multiple different capture methods. For example, a base point may be first determined in the breakpoint region, and then the base point may be used as the center point to capture multiple regional sub-images of different sizes in the target image. The base point may be the center point of the breakpoint region or a random pixel point in the breakpoint region. Before capturing the regional sub-images, the size of each regional sub-image may be predetermined. In practical applications, the size of the regional sub-image may be gradually increased. For example, in Figure 6 In the example, the length and width of the smallest region sub-image can be set to 2 / 16 of the length and width of the target image, and the ratio of the length and width of the subsequent region sub-images to the length and width of the target image can be gradually increased from 2 / 16 to 2 / 2. In this way, region sub-images of different sizes can be cut out. The size of the region sub-image determined in this way can be used as the theoretical size.

[0052] Of course, since the region sub-image is intercepted from the target image, the area covered by the final region sub-image will not exceed the area of ​​the target image, so the actual size of the final intercepted region sub-image will have a certain deviation from the above theoretical size. The final intercepted region sub-image is actually the intersection between the area covered by the above theoretical size and the area of ​​the target image.

[0053] In one embodiment, different base points can be selected, and then different base points are used to intercept regional sub-images of different sizes, thereby enriching the diversity of features in the regional sub-image. Specifically, a base point sequence can be determined in the breakpoint region, and the base point sequence includes multiple target base points. Among them, when selecting each target base point in the breakpoint region, different sampling probabilities can be used to perform random sampling in the breakpoint region, and the multiple randomly sampled pixel points are used as each target base point in the base point sequence. Subsequently, for each target base point, the interception size corresponding to the target base point can be determined. The interception size can be determined in a similar manner to the size of the regional sub-image in the aforementioned embodiment. Subsequently, the target base point can be used as the center point to intercept a regional sub-image in the target image that meets the interception size corresponding to the target base point. For example, the base point sequence includes 5 target base points, and these 5 target base points correspond to different interception sizes. In this way, when intercepting a regional sub-image from the target image, these 5 target base points can be used as center points to intercept a regional sub-image of corresponding size from the target image. The size of the region sub-image determined in this way can be used as the theoretical size. Similarly, the region sub-image finally intercepted is actually the intersection between the area covered by the theoretical size and the area of ​​the target image.

[0054] S5: identifying the blind path occupancy prediction results represented by the respective area sub-images, and judging whether the blind path area in the target image is occupied according to the blind path occupancy prediction results.

[0055] In this embodiment, after a plurality of region sub-images intersecting with the breakpoint region are captured from the target image, the blind path occupancy prediction results represented by the region sub-images can be determined one by one. The blind path occupancy prediction results can indicate that the blind path is occupied or not occupied.

[0056] In practical applications, considering that the blind path occupancy prediction results usually only include two types of results (occupied or unoccupied), an image classification model can be used to process the regional sub-image to output the blind path occupancy prediction results. Specifically, during the training process of the image classification model, a classification image sample can be first obtained, and the classification image sample can have a marked classification label. Among them, the classification image sample can include a breakpoint area in the blind path area, which may be caused by the occupation of the blind path area or the shooting angle of the classification image sample. Based on whether the blind path area in the classification image sample is occupied, a corresponding classification label can be set. The classification image sample is processed by the image classification model to obtain a predicted classification result. The error information between the predicted classification result and the standard classification result represented by the classification label can be used to correct the image classification model. Among them, the image classification model can be implemented using network architectures such as AlexNet, ResNet, and VGG. The error information involved in the training process of the image classification model can be represented by cross entropy or other error functions.

[0057] In this implementation, after the image classification model is trained, each region sub-image may be processed using the image classification model, thereby outputting a blind path occupancy prediction result represented by each region sub-image.

[0058] In this embodiment, different blind path occupancy prediction results can be counted to determine whether the blind path area in the target image is occupied. Specifically, in one embodiment, the number of first results representing that the blind path is occupied and the number of second results representing that the blind path is not occupied can be counted from each of the blind path occupancy prediction results. If the number of the first results is greater than the number of the second results, it is determined that the blind path area in the target image is occupied; if the number of the first results is less than the number of the second results, it is determined that the blind path area in the target image is not occupied.

[0059] In another embodiment, considering the different sizes of regional sub-graphs, the confidence of the corresponding blind path occupancy prediction results may also be different. For example, for a regional sub-graph with a larger size, the confidence of the corresponding blind path occupancy prediction result may be higher. For another example, the confidence of the blind path occupancy prediction result corresponding to the regional sub-graph may be normally distributed with the size of the regional sub-graph. For regional sub-graphs with smaller or larger sizes, the corresponding confidence is relatively low, while for regional sub-graphs with a size in the middle area, the corresponding confidence is relatively high.

[0060] In view of this, respective weight values ​​can be assigned in advance to the blind path occupancy prediction results of each regional sub-image. The higher the confidence of the regional sub-image, the larger the assigned weight value can be. Subsequently, according to the assigned weight values, the first weight sum representing the occupation of the blind path and the second weight sum representing the unoccupied blind path can be statistically calculated in each of the blind path occupancy prediction results. If the first weight sum is greater than the second weight sum, it is determined that the blind path area in the target image is occupied; if the first weight sum is less than the second weight sum, it is determined that the blind path area in the target image is not occupied.

[0061] The technical solution provided by one or more embodiments of the present disclosure does not need to pay attention to the type of obstacles occupying the blind path, but only detects the breakpoint area in the blind path area, and based on the breakpoint area, screenshots of multiple different area sub-graphs are taken. Subsequently, by analyzing the blind path occupancy prediction results represented by each area sub-graph, it can be comprehensively determined whether the current blind path area is occupied.

[0062] The above-mentioned breakpoint area can reflect the discontinuity of the blind path area. When the blind path area in the target image shows discontinuity characteristics, it indicates that there may be obstacles in the blind path area. By analyzing the breakpoint area, the actual type of obstacles in the blind path area can be ignored, making the technical solution of the present disclosure more universal in blind path occupancy detection. At the same time, the above process can automatically process the target image and output the final judgment result, thereby improving the efficiency of blind path occupancy detection.

[0063] See also Figure 7 In one embodiment of the present disclosure, a blind path occupancy detection device is provided, the device comprising:

[0064] The area recognition unit 100 is used to obtain a target image to be detected and recognize a blind path area in the target image;

[0065] A sub-image capture unit 200 is used to detect a breakpoint region located in the blind path region in the target image, and based on the breakpoint region, capture a plurality of region sub-images from the target image, wherein each of the region sub-images has an intersection with the breakpoint region;

[0066] The judging unit 300 is configured to identify the blind path occupancy prediction results represented by the respective area sub-images, and judge whether the blind path area in the target image is occupied according to the blind path occupancy prediction results.

[0067] Among them, the specific processing logic of each functional module can be found in the description of the aforementioned method implementation method, which will not be repeated here.

[0068] Each unit described in the above embodiments may be implemented by a computer chip or a product having a certain function. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0069] For the convenience of description, the above devices are described in terms of functions and are described separately in various units. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0070] See also Figure 8 An embodiment of the present disclosure further provides an electronic device, which includes a memory and a processor, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the above-mentioned blind path occupancy detection method is implemented.

[0071] The present disclosure also provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the above-mentioned blind path occupancy detection method is implemented.

[0072] The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0073] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory, that is, implementing the method in the above method embodiment.

[0074] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0075] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0076] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the embodiments of the device, equipment, and storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0077] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

[0078] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for detecting blind path occupancy, characterized in that: The method comprises: Acquire a target image to be detected, and identify a blind path area in the target image; Detecting a breakpoint region located in the blind path region in the target image, and based on the breakpoint region, intercepting a plurality of region sub-images from the target image, wherein each of the region sub-images has an intersection with the breakpoint region; The blind path occupancy prediction results represented by the respective area sub-images are identified, and according to the blind path occupancy prediction results, it is determined whether the blind path area in the target image is occupied.

2. The method according to claim 1, characterized in that: The target image is processed by a blind path segmentation model to obtain a blind path region recognition result of the target image, wherein the blind path region recognition result includes one or more recognized blind path regions; The blind road segmentation model is trained in the following way: Acquire a blind path image sample, wherein the blind path image sample has a binary image label, and the binary image label distinguishes a blind path area and a non-blind path area in the blind path image sample by different pixel values; Processing the blind path image sample by a blind path segmentation model to generate a predicted binary image corresponding to the blind path image sample; Generate error information between the predicted binary image and the binary image label, and correct the blind path segmentation model based on the error information.

3. The method according to claim 1, characterized in that After identifying the blind road area in the target image, the method further includes: For any target blind path area in the blind path area recognition result, determine the area ratio of the target blind path area in the target image; If the area proportion is less than or equal to the specified proportion threshold, the target blind path area is eliminated from the blind path area recognition result; if the area proportion is greater than the specified proportion threshold, the target blind path area is retained in the blind path area recognition result.

4. The method according to claim 1, characterized in that Detecting the breakpoint area in the target image within the blind path area includes: Processing the target image through a breakpoint detection model to obtain a breakpoint prediction result corresponding to the target image, wherein the breakpoint prediction result is used to characterize the position of the candidate breakpoint in the target image; The candidate breakpoints located outside the blind path area are removed from the breakpoint prediction result, and the area formed by the candidate breakpoints retained in the breakpoint prediction result is used as the breakpoint area obtained by detection.

5. The method according to claim 1 or 4, characterized in that: Based on the breakpoint area, extracting a plurality of area sub-images from the target image comprises: Determine a base point in the breakpoint area, and use the base point as a center point to intercept a plurality of area sub-images of different sizes in the target image; or A base point sequence is determined in the breakpoint area, wherein the base point sequence includes a plurality of target base points; for any target base point, a cutout size corresponding to the target base point is determined, and a region sub-image conforming to the cutout size is cutout in the target image with the target base point as the center point.

6. The method according to claim 1, characterized in that Determining whether the blind path area in the target image is occupied includes: A first number of results indicating that a blind path is occupied and a second number of results indicating that a blind path is not occupied are counted from each of the blind path occupancy prediction results, wherein if the first number of results is greater than the second number of results, it is determined that the blind path area in the target image is occupied; if the first number of results is less than the second number of results, it is determined that the blind path area in the target image is not occupied.

7. The method according to claim 1, characterized in that Determining whether the blind path area in the target image is occupied includes: Assigning respective weight values ​​to the respective blind path occupation prediction results, and according to the assigned weight values, statistically calculating the sum of first weights representing that the blind paths are occupied, and statistically calculating the sum of second weights representing that the blind paths are not occupied, in the respective blind path occupation prediction results; If the first weight sum is greater than the second weight sum, it is determined that the blind path area in the target image is occupied; if the first weight sum is less than the second weight sum, it is determined that the blind path area in the target image is not occupied.

8. A blind path occupancy detection device, characterized in that: The device comprises: A region recognition unit, used for acquiring a target image to be detected and identifying a blind path region in the target image; A sub-image capture unit, configured to detect a breakpoint region located in the blind path region in the target image, and capture a plurality of region sub-images from the target image based on the breakpoint region, wherein each of the region sub-images has an intersection with the breakpoint region; A judging unit is used to identify the blind path occupancy prediction results represented by the respective area sub-images, and judge whether the blind path area in the target image is occupied according to the blind path occupancy prediction results.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images