Training methods for blind spot pedestrian detection models, blind spot pedestrian detection methods and devices

By converting the vehicle blind spot image set from the first format to the second format and converting the annotation information, a blind spot pedestrian detection model with a smaller data volume is trained, which solves the problem of excessive detection time caused by the limited computing power of the vehicle terminal, reduces the false alarm rate and false negative rate, and improves the detection accuracy.

CN117612131BActive Publication Date: 2026-11-17CHINA AUTOMOTIVE INNOVATION CORP
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311379594.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-10-23
Publication Date
2026-11-17
Estimated Expiration
2043-10-23

AI Technical Summary

Technical Problem

In existing technologies, the computing power of vehicle terminals is limited, and the pedestrian detection model based on deep learning takes too long, resulting in an increase in the false alarm rate and false negative rate of pedestrian detection in blind spots.

Method used

By converting the original first-format vehicle blind spot image set into the second format and converting pedestrian annotation information, a blind spot pedestrian detection model with a smaller data volume is trained. The target image detection model is then trained using a selected image training set to improve detection accuracy.

Benefits of technology

It reduces the data processing volume and time for pedestrian detection in blind spots, reduces frame loss, lowers the false alarm rate and false negative rate, and improves the accuracy of pedestrian warning in blind spots.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117612131B_ABST
    Figure CN117612131B_ABST
Patent Text Reader

Abstract

This application discloses a training method, a blind spot pedestrian detection method, and an apparatus for a blind spot pedestrian detection model, belonging to the field of computer technology. In the technical solution provided by this application, the original first-format vehicle blind spot image set is converted to a second format for pedestrian annotation. Then, the annotation information of the second-format image is converted to obtain the annotation information under the first-format image. The image processing model is trained directly using the smaller first-format vehicle blind spot image set and the converted pedestrian annotation information, resulting in a blind spot pedestrian detection model with fewer model parameters. Furthermore, the first-format images detected by this model are relatively small. Therefore, the data processing volume and time during pedestrian detection by the blind spot pedestrian detection model can be reduced, frame loss can be decreased, thereby reducing the false alarm rate and false negative rate of blind spot pedestrian detection and improving the accuracy of blind spot pedestrian warning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a training method for a blind spot pedestrian detection model, a blind spot pedestrian detection method, and an apparatus. Background Technology

[0002] As the electrification of the automotive industry accelerates, more and more cars with driver assistance features are entering households. At the same time, vehicle accidents caused by dangerous driving behaviors are rising year by year, raising urgent concerns about the safety of driver assistance features during driving.

[0003] In existing technologies, the computing power of vehicle terminals is limited. The pedestrian detection process based on deep learning is too time-consuming, resulting in severe frame loss and an increase in the false alarm rate and false negative rate of pedestrian detection in blind spots. Summary of the Invention

[0004] This application provides a training method, a blind spot pedestrian detection method, and an apparatus for a blind spot pedestrian detection model. These methods can reduce the amount of data processing and the data processing time when the blind spot pedestrian detection model performs pedestrian detection, reduce frame loss, and thus reduce the false alarm rate and false negative rate of blind spot pedestrian detection.

[0005] According to one aspect of the embodiments of this application, a method for training a blind spot pedestrian detection model is provided, the method comprising: Acquire a first vehicle blind spot image set, wherein the image format corresponding to the first vehicle blind spot image set is a first format; The images in the first vehicle blind spot image set are converted to a second vehicle blind spot image set. The image format corresponding to the second vehicle blind spot image set is a second format. The image data volume corresponding to the first format is less than the image data volume corresponding to the second format. Obtain the first pedestrian annotation information corresponding to the second vehicle blind spot image set; The first pedestrian annotation information is converted to obtain the second pedestrian annotation information corresponding to the first vehicle blind spot image set; The image processing model is trained based on the first vehicle blind spot image set and the second pedestrian annotation information to obtain the blind spot pedestrian detection model.

[0006] In an exemplary embodiment, the first pedestrian labeling information includes first label information and second label information. The first label information is used to label pedestrians who are not obscured, and the second label information is used to label pedestrians who are obscured by an obstruction.

[0007] In an exemplary embodiment, the obstruction includes at least one of a circular column or a rod-shaped object.

[0008] In an exemplary embodiment, the first pedestrian annotation information includes first pedestrian coordinate information; the step of converting the first pedestrian annotation information to obtain the second pedestrian annotation information corresponding to the first vehicle blind spot image set includes: The first pedestrian coordinate information is transformed to obtain the second pedestrian coordinate information corresponding to the first vehicle blind spot image set; Based on the label information corresponding to the first pedestrian coordinate information, determine the label information corresponding to the second pedestrian coordinate information; The second pedestrian annotation information is obtained based on the second pedestrian coordinate information and the corresponding label information.

[0009] In an exemplary embodiment, the method further includes: Acquire a third vehicle blind spot image set and pedestrian detection information corresponding to the third vehicle blind spot image set; The step of training a preset image processing model based on the first vehicle blind spot image set and the second pedestrian annotation information to obtain a blind spot pedestrian detection model includes: The first vehicle blind spot image set is filtered based on the second pedestrian marking information, and the third vehicle blind spot image set is filtered based on the pedestrian detection information to obtain a filtered image training set. The filtered image training set includes vehicle blind spot images corresponding to at least two pedestrian feature conditions. The target image detection model is trained based on the selected image training set to obtain the blind spot pedestrian detection model.

[0010] In an exemplary embodiment, the step of filtering the first vehicle blind spot image set based on the second pedestrian marking information and filtering the third vehicle blind spot image set based on the pedestrian detection information to obtain a filtered image training set includes: When the number of images in the first vehicle blind spot image set reaches a first quantity threshold, images in the first vehicle blind spot image set that meet the first pedestrian feature condition are selected to obtain the initial selected image training set. The preset image detection model is trained based on the initial selected image training set to obtain the intermediate blind spot pedestrian detection model. The pedestrian detection information is obtained by the intermediate blind spot pedestrian detection model detecting the third vehicle blind spot image set. Based on the pedestrian detection information, target images that meet the second pedestrian feature conditions in the third vehicle blind spot image set are added to the filtered image training set to obtain the updated filtered image training set. The step of training the target image detection model based on the selected image training set to obtain the blind spot pedestrian detection model includes: The intermediate blind spot pedestrian detection model is trained based on the updated selected image training set to obtain the blind spot pedestrian detection model.

[0011] In an exemplary embodiment, obtaining the third vehicle blind spot image set and the pedestrian detection information corresponding to the third vehicle blind spot image set includes: Obtain the blind spot image set of the third vehicle; Based on the aforementioned pedestrian detection model for the intermediate blind spot, pedestrian detection is performed on the images in the third vehicle blind spot image set to obtain the pedestrian detection information.

[0012] In an exemplary embodiment, after performing pedestrian detection on the images in the third vehicle blind spot image set based on the intermediate blind spot pedestrian detection model to obtain the pedestrian detection information, the method further includes: Determine whether the pedestrian detection information meets the training constraints; If the pedestrian detection information does not meet the training constraints, determine the number of target images; If the number of target images is less than a preset threshold, the second pedestrian feature condition is updated, and the step of adding the target images that meet the second pedestrian feature condition in the third vehicle blind spot image set to the filtered image training set based on the pedestrian detection information is performed to obtain the updated filtered image training set.

[0013] In an exemplary embodiment, after determining whether the pedestrian detection information meets the training constraints, the method further includes: If the pedestrian detection information meets the training constraints, the current intermediate blind spot pedestrian detection model is determined as the blind spot pedestrian detection model.

[0014] In an exemplary embodiment, the method further includes: Obtain anomaly detection video; Based on the vehicle blind spot pedestrian detection model, pedestrian detection is performed on the abnormal detection video to obtain an image with abnormal detection results; The selected image training set is updated based on the images with abnormal detection results; The vehicle blind spot pedestrian detection model is trained based on the updated selected image training set to obtain the updated vehicle blind spot pedestrian detection model.

[0015] In an exemplary embodiment, the first format is YUV format and the second format is RGB format.

[0016] According to one aspect of the embodiments of this application, a method for pedestrian detection in blind spots is provided, the method comprising: The vehicle blind spot image in the first format is input into the blind spot pedestrian detection model for pedestrian detection to obtain the blind spot pedestrian detection result; the blind spot pedestrian detection model is a blind spot pedestrian detection model trained by the method described above; If the pedestrian detection result in the blind spot indicates that a pedestrian exists in the vehicle's blind spot image, a prompt message is output.

[0017] In an exemplary embodiment, the blind spot pedestrian detection result includes the lateral distance from the pedestrian to the vehicle's blind spot. When the blind spot pedestrian detection result indicates the presence of a pedestrian within the blind spot area, the output of a prompt message includes: If the lateral distance is less than or equal to the first distance threshold, output a level one alarm message; If the lateral distance is less than or equal to the second distance threshold, output a level 2 alarm message; If the lateral distance is less than or equal to the third distance threshold, a level three alarm message is output.

[0018] According to one aspect of the embodiments of this application, a training apparatus for a blind spot pedestrian detection model is provided, the apparatus comprising: The image acquisition module is used to acquire a first vehicle blind spot image set, wherein the image format corresponding to the first vehicle blind spot image set is a first format; The format conversion module is used to convert the format of the images in the first vehicle blind spot image set to obtain the second vehicle blind spot image set. The image format corresponding to the second vehicle blind spot image set is the second format, and the image data volume corresponding to the first format is less than the image data volume corresponding to the second format. The annotation information acquisition module is used to acquire the first pedestrian annotation information corresponding to the second vehicle blind spot image set; The annotation information conversion module is used to convert the first pedestrian annotation information to obtain the second pedestrian annotation information corresponding to the first vehicle blind spot image set; The model training module is used to train a preset image processing model based on the first vehicle blind spot image set and the second pedestrian annotation information to obtain a blind spot pedestrian detection model.

[0019] According to one aspect of the embodiments of this application, a blind spot pedestrian detection device is provided, the device comprising: The image detection module is used to input the vehicle blind spot image in the first format into the blind spot pedestrian detection model for pedestrian detection, and obtain the blind spot pedestrian detection result; the blind spot pedestrian detection model is the blind spot pedestrian detection model trained by the above method; The information output module is used to output a prompt message when the pedestrian detection result in the blind spot indicates that a pedestrian exists in the vehicle's blind spot image.

[0020] According to one aspect of the embodiments of this application, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, the at least one program, the code set or instruction set being loaded and executed by the processor to implement the above-described training method or blind spot pedestrian detection method for the blind spot pedestrian detection model.

[0021] According to one aspect of the embodiments of this application, a computer-readable storage medium is provided, wherein at least one instruction, at least one program, code set, or instruction set is stored in the storage medium, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the above-described training method or blind spot pedestrian detection method for the blind spot pedestrian detection model.

[0022] According to one aspect of the embodiments of this application, a computer program product is provided, the computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, causing the computer device to perform a training method or a blind spot pedestrian detection method for the above-described blind spot pedestrian detection model.

[0023] The technical solution provided in this application can bring the following beneficial effects: After converting the original first-format vehicle blind spot image set into the second format for pedestrian annotation, the annotation information of the second-format images is then converted to obtain the annotation information under the first-format images. By directly training the image processing model with the smaller first-format vehicle blind spot image set and the converted pedestrian annotation information, a blind spot pedestrian detection model with fewer model parameters can be obtained. Furthermore, the first-format images detected by this model are relatively small. Therefore, the amount of data processing and the data processing time when the blind spot pedestrian detection model performs pedestrian detection can be reduced, frame loss can be reduced, thereby reducing the false alarm rate and false negative rate of blind spot pedestrian detection and improving the accuracy of blind spot pedestrian warning.

[0024] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of a training method for a blind spot pedestrian detection model provided in one embodiment of this application; Figure 2 This is a flowchart of a blind spot pedestrian detection method provided in one embodiment of this application; Figure 3 A flowchart of a dangerous driving behavior early warning method based on a single-channel deep learning model is shown; Figure 4 This is a block diagram of a training device for a blind spot pedestrian detection model provided in one embodiment of this application; Figure 5 A block diagram of a blind spot pedestrian detection device provided in one embodiment of this application; Figure 6 This is a structural block diagram of a computer device provided in one embodiment of this application. Detailed Implementation

[0027] Please refer to Figure 1 The diagram illustrates a flowchart of a training method for a blind spot pedestrian detection model according to an embodiment of this application. This method can be applied to a computer device, which refers to an electronic device capable of data computation and processing. The method may include the following steps (110-150).

[0028] Step 110: Obtain the first vehicle blind spot image set.

[0029] Optionally, the image format corresponding to the first vehicle blind spot image set is a first format. Specifically, the first format can be YUV format.

[0030] In one embodiment, the device acquires vehicle blind spot driving video captured by a camera in the vehicle; and generates a first vehicle blind spot image set based on the vehicle blind spot driving video.

[0031] Specifically, the aforementioned camera can be a monocular camera. The monocular camera is used to collect driving videos of the vehicle's blind spots while the vehicle is in motion. Through the aforementioned video, an original three-channel YUV-NV12 format (the first format can be YUV-NV12 format) image frame sequence can be obtained, thus obtaining the aforementioned first vehicle blind spot image set.

[0032] Step 120: Convert the format of the images in the first vehicle blind spot image set to obtain the second vehicle blind spot image set.

[0033] Optionally, the image format corresponding to the second vehicle blind spot image set is a second format. The image data size corresponding to the first format is smaller than that corresponding to the second format. Specifically, the second format can be RGB format.

[0034] Specifically, the images in the first vehicle blind spot image set in YUV-NV12 format can be converted into corresponding RGB format images to obtain the second vehicle blind spot image set.

[0035] Step 130: Obtain the first pedestrian annotation information corresponding to the second vehicle blind spot image set.

[0036] The first pedestrian annotation information is obtained by annotating pedestrians in the vehicle blind spot image set of the second vehicle blind spot image set using an annotation tool. This first pedestrian annotation information includes pedestrian coordinate information and pedestrian label information corresponding to each image in the second vehicle blind spot image set. The pedestrian coordinate information can be the bounding box corresponding to the pedestrian in the image, and the pedestrian label information can be the label corresponding to the pedestrian.

[0037] Optionally, the first pedestrian labeling information includes first label information and second label information. The first label information is used to label pedestrians who are not obscured, and the second label information is used to label pedestrians who are obscured by an obstruction. Optionally, the obstruction includes at least one of a circular pillar or a pole-shaped object.

[0038] In practical applications, when using annotation tools, pedestrian annotations in RGB image samples can be divided into two categories: person and person_hard. Clear and complete pedestrians (i.e., pedestrians that are not obscured as mentioned above) are labeled as person (i.e., the first label information mentioned above); pedestrians that are obscured by circular pillars or pole-shaped objects (i.e., the obscured objects mentioned above) are labeled as person_hard (i.e., the second label information mentioned above).

[0039] The reason for this labeling is that if pedestrians obscured by circular pillars or poles are also labeled as pedestrians, the model will learn the texture information of the circular pillars or poles. This increases the probability of vehicles misidentifying circular pillars or poles as pedestrians while in motion. This is especially true when vehicles equipped with driver assistance systems are continuously traveling on the road, generating a massive number of samples per vehicle per day. A high proportion of these samples contain circular or pole-shaped traffic signs. If vehicles frequently misidentify these as pedestrians, the driver's experience will be poor. The technical solution provided in this application, by separating pedestrian labeling into the aforementioned two types of information, allows the model to learn more detailed pedestrian features, thereby improving the detection accuracy of pedestrians in blind spots.

[0040] Step 140: Convert the first pedestrian annotation information to obtain the second pedestrian annotation information corresponding to the first vehicle blind spot image set.

[0041] By using the annotation tool described above, the coordinate set of the pedestrian in the image coordinate system within the RGB three-channel image training set (i.e., the aforementioned second vehicle blind spot image set) can be obtained, which is the coordinate information of the first pedestrian. The aforementioned first pedestrian annotation information includes the aforementioned first pedestrian coordinate information. Accordingly, step 140 includes the following sub-steps: The first pedestrian coordinate information is transformed to obtain the second pedestrian coordinate information corresponding to the first vehicle blind spot image set. Specifically, the coordinate set of pedestrians in the RGB three-channel image training set (i.e., the first pedestrian coordinate information mentioned above) can be converted into the coordinate set of pedestrians in the original three-channel YUV-NV12 image training set, i.e., the second pedestrian annotation information mentioned above.

[0042] Based on the label information corresponding to the first pedestrian coordinate information, the label information corresponding to the second pedestrian coordinate information is determined. There is a one-to-one correspondence between the first and second pedestrian coordinate information, and the label information corresponding to the first pedestrian coordinate information can be used as the label information corresponding to the second pedestrian coordinate information according to this correspondence.

[0043] Based on the second pedestrian coordinate information and the corresponding label information, the second pedestrian annotation information is obtained. After mapping the second pedestrian coordinate information and the corresponding label information, the second pedestrian annotation information corresponding to the first vehicle blind spot image set can be obtained. This includes the coordinate information and label information of the pedestrians in the first vehicle blind spot image set, and the label information includes the two types of label information mentioned above.

[0044] This conversion allows for accurate acquisition of the annotation information corresponding to the first vehicle blind spot image set, reducing the difficulty of image labeling. It enables the model to be trained directly on the first vehicle blind spot image set with a smaller data volume based on the converted annotation information, thereby reducing the number of model parameters.

[0045] Step 150: Train the preset image processing model based on the first vehicle blind spot image set and the second pedestrian annotation information to obtain the blind spot pedestrian detection model.

[0046] After inputting the aforementioned first vehicle blind spot image set into a preset image processing model, detection results can be obtained. These results are then compared with the second pedestrian annotation information, allowing for iterative training of the image processing model until the matching degree between the model's output detection results and the second pedestrian annotation information meets the requirements. This yields the blind spot pedestrian detection model. This model can detect pedestrians in the vehicle's blind spot video while the vehicle is in motion, thereby determining whether a pedestrian is present within the vehicle's blind spot.

[0047] In the technical solution provided in this application embodiment, the original vehicle blind spot image set in the first format is converted into the second format and then pedestrian annotation is performed. The annotation information of the second format image is then converted to obtain the annotation information under the first format image. The image processing model is trained directly using the vehicle blind spot image set in the first format with a smaller data volume and the converted pedestrian annotation information. This results in a blind spot pedestrian detection model with a smaller number of model parameters. Furthermore, the first format image detected by this model is relatively small. Therefore, the data processing volume and data processing time when the blind spot pedestrian detection model performs pedestrian detection can be reduced, frame loss can be reduced, thereby reducing the false alarm rate and false negative rate of blind spot pedestrian detection and improving the accuracy of blind spot pedestrian warning.

[0048] In an exemplary embodiment, the training method further includes: acquiring a third vehicle blind spot image set and pedestrian detection information corresponding to the third vehicle blind spot image set. The third vehicle blind spot image set may be a training set composed of blind spot images collected during vehicle operation on the road, and the pedestrian detection information may be detection information obtained by a relevant model performing pedestrian detection on the images. The relevant model here may be the blind spot pedestrian detection model trained in this embodiment, or other pedestrian detection models; this embodiment does not limit the specific model used.

[0049] Accordingly, the specific implementation process of step 150 above can be as follows: The first vehicle blind spot image set is filtered based on second pedestrian marking information, and the third vehicle blind spot image set is filtered based on pedestrian detection information to obtain a filtered image training set. Optionally, the filtered image training set includes vehicle blind spot images corresponding to at least two pedestrian feature conditions. The aforementioned pedestrian feature conditions are used to represent pedestrian features, and different feature conditions can correspond to pedestrians with different features. The aforementioned pedestrian feature conditions can serve as the basis for filtering the two image sets.

[0050] The target image detection model is trained based on the selected image training set to obtain the blind spot pedestrian detection model.

[0051] The aforementioned first and third vehicle blind spot image sets are mostly images collected when vehicles are driving on the road. Most of the scenes in the image sets are repetitive and the number of samples accounts for a large part of the total training set. However, the number of samples in special scenarios is small. When these few special samples are added to the training set, the proportion of these special samples in the total training set is small. These few special samples have little impact on the training results of the deep learning model, such as the model's accuracy, precision, false positive rate, and false negative rate. This can easily lead to the problem that the model has high accuracy on the preset test set, but low accuracy in detecting special samples.

[0052] Therefore, in the technical solution provided in this application embodiment, the first vehicle blind spot image set is filtered to obtain a filtered image training set with rich pedestrian features and a controllable number. The range of filtered images is increased, not only the first vehicle blind spot image set is filtered, but also the third vehicle blind spot image set is filtered, and the images filtered from the third vehicle blind spot image set are also added to the filtered image training set. This improves the feature richness of the filtered image training set, increases the proportion of special samples in the dataset, thereby helping to improve the model's detection accuracy for special samples and ensuring the overall detection accuracy of the model.

[0053] In one possible implementation, the number of images in the first vehicle blind spot image set may be continuously updated, and the process of filtering the first vehicle blind spot image set based on the second pedestrian marking information may be as follows: When the number of images in the first vehicle blind spot image set reaches a first threshold, images that meet the first pedestrian feature condition are selected from the first vehicle blind spot image set to obtain an initial selected image training set. Then, a preset image detection model is trained based on the initial selected image training set to obtain an intermediate blind spot pedestrian detection model. For example, when the number of images in the first vehicle blind spot image set reaches millions, the trained model cannot learn the features of a small number of samples in a specific scene, so it is necessary to simplify the model training set and remove redundant data. At this time, based on the second pedestrian marking information, 30,000 sample images with pedestrian annotation aspect ratios between 2.0 and 2.5 (first pedestrian feature condition) from the original three-channel YUV-NV12 image training set (first vehicle blind spot image set) can be used as the initial training set, i.e., the initial selected image training set. Then, the initial model is trained using the initial training set, i.e., the aforementioned intermediate blind spot pedestrian detection model. The aforementioned intermediate blind spot pedestrian detection model refers to the blind spot pedestrian detection model trained in the intermediate step of the training process.

[0054] After training the pedestrian detection model for the middle blind spot, pedestrian detection can be performed on the images in the third vehicle blind spot image set based on the middle blind spot pedestrian detection model to obtain pedestrian detection information. Therefore, the pedestrian detection information corresponding to the aforementioned third vehicle blind spot image set can be obtained by the middle blind spot pedestrian detection model detecting pedestrians in the third vehicle blind spot image set.

[0055] Accordingly, the process of filtering the blind spot image set of third vehicles based on pedestrian detection information can be as follows: Based on pedestrian detection information, target images that meet the second pedestrian feature conditions in the third vehicle blind spot image set are added to the filtered image training set to obtain an updated filtered image training set; then, the intermediate blind spot pedestrian detection model is trained based on the updated filtered image training set to obtain the blind spot pedestrian detection model.

[0056] Using the aforementioned third vehicle blind spot image set, the middle blind spot pedestrian detection model can be preliminarily validated. The pedestrian detection information detected by the middle blind spot pedestrian detection model can be used to filter out target images that the middle blind spot pedestrian detection model cannot accurately identify (which can be understood as the pedestrian features in these images have not yet been learned by the middle blind spot pedestrian detection model), namely, the target images that meet the second pedestrian feature condition. Adding these target images to the filtered image training set can improve the richness of pedestrian features in the filtered image training set. Then, using the updated filtered image training set to iteratively train the middle blind spot pedestrian detection model, the model can learn more pedestrian features. Furthermore, the filtered image training set still retains the original training images. While the model learns more pedestrian features, it will not lose the original features, thereby improving the detection accuracy and comprehensiveness of the vehicle blind spot pedestrian detection model.

[0057] Optionally, the conditional dimensions of the second pedestrian feature condition mentioned above include the pedestrian recognition result (indicating whether a pedestrian is present or not), the pedestrian aspect ratio range, and the detection confidence threshold. The pedestrian aspect ratio mentioned above is the aspect ratio of the pedestrian bounding box detected by the pedestrian detection model in the middle blind zone, and the detection confidence threshold represents the probability that the pedestrian bounding box contains a pedestrian.

[0058] To facilitate understanding, an example is provided below. The aforementioned third vehicle blind spot image set could be a dataset of 1 million images. The intermediate blind spot pedestrian detection model can detect pedestrians in the images of the third vehicle blind spot image set, thereby obtaining pedestrian detection information for each image, including the corresponding pedestrian bounding box (if any) and the corresponding detection confidence score. When selecting samples, samples with pedestrian aspect ratios between 2.0 and 2.5 and model scores (i.e., the aforementioned detection confidence scores) below 0.5 (the confidence threshold) (meaning these samples have low detection accuracy) can be added to the initial selected image training set for training.

[0059] Optionally, the number of samples added to the selected image training set each time is 10% of the selected image training set (or other values, which are not limited here), that is, 3000 samples are added to the initial training set. Before adding to the initial training set, it is necessary to check whether there are samples with unlabeled pedestrians mixed in, to avoid samples with unlabeled pedestrians entering the selected image training set. The presence of unlabeled pedestrians can also be filtered using a second pedestrian feature condition.

[0060] In another possible implementation, after detecting pedestrian detection information, it is also determined whether the pedestrian detection information meets the training constraints. If the pedestrian detection information does not meet the training constraints, determine the number of target images; if it does not meet the training constraints, it means that the model has not yet been trained to the final version.

[0061] If the number of target images is less than a preset threshold, the second pedestrian feature condition is updated, and the target images that meet the second pedestrian feature condition in the third vehicle blind spot image set are added to the filtered image training set based on pedestrian detection information to obtain the updated filtered image training set.

[0062] If the number of target images is less than the preset threshold, it means that the model can detect and identify the images corresponding to the current second pedestrian feature conditions relatively accurately. Therefore, the second pedestrian feature conditions can be updated. Specifically, the aspect ratio range of pedestrians and the detection confidence threshold can be updated. New target images are selected and added to the selected image training set using the updated second pedestrian feature conditions, thereby enriching the feature dimensions of the selected image training set layer by layer and improving the detection accuracy of the model.

[0063] For example, if the detection confidence threshold is set to 0.5 and the pedestrian aspect ratio is between 2.0 and 2.5, and the number of selected target images is less than 1000, then the detection confidence threshold is set to 0.6 and the aspect ratio is updated to between 1.5 and 2.0, and the selection is repeated. This process is repeated, using detection confidence thresholds of 0.6, 0.7, 0.8, and 0.9 combined with the pedestrian aspect ratio range of 1.5 to 2.5 to select samples. Furthermore, before adding samples to the training set, it is checked whether any unlabeled pedestrian samples are mixed in.

[0064] If the number of target images is greater than or equal to a preset threshold, keep the current second pedestrian feature condition unchanged, and continue to execute the step of adding the target images that meet the second pedestrian feature condition in the third vehicle blind spot image set to the filtered image training set based on pedestrian detection information, so as to obtain an updated filtered image training set.

[0065] If the number of target images is greater than or equal to the preset threshold, it means that the model has not yet accurately detected and identified the images corresponding to the current second pedestrian feature condition, and it needs to continue iterative training under the current second pedestrian feature condition.

[0066] If the pedestrian detection information meets the training constraints, the current intermediate blind zone pedestrian detection model can be identified as the blind zone pedestrian detection model. If the model meets the preset training conditions, it can be considered to have met the preset requirements and can be used as the final version of the blind zone pedestrian detection model, thus obtaining a blind zone pedestrian detection model with higher accuracy. These training constraints can include reaching the maximum number of iterations, passing validation on the test set, and achieving preset accuracy in the selected image training set, etc.

[0067] In another possible implementation, the above training method further includes the following steps: Acquire anomaly detection videos; perform pedestrian detection on the anomaly detection videos based on the vehicle blind spot pedestrian detection model to obtain images with abnormal detection results; update the filtered image training set based on the images with abnormal detection results; train the vehicle blind spot pedestrian detection model based on the updated filtered image training set to obtain the updated vehicle blind spot pedestrian detection model.

[0068] The aforementioned anomaly detection videos can be those from historical detection processes where pedestrian detection results in the blind spot were incorrect. Examples include false positive videos (detecting a pedestrian even when no one is present) and missed detection videos (detecting a pedestrian even when no one is present). Inputting these videos back into the vehicle blind spot pedestrian detection model yields the latest detection results. Images with still abnormal detection results are then added to the filtered image training set, enriching the feature dimensions of the training set. Based on this updated filtered image training set, the vehicle blind spot pedestrian detection model is further trained, resulting in a more accurate model. Before adding missed and false positive samples to the training set, it's necessary to check if the added data contains unlabeled samples to improve model accuracy and provide a final, stable blind spot pedestrian detection model.

[0069] Optionally, the video image frame samples detected in the anomaly detection video can also be added to the aforementioned third vehicle blind spot image set for later iteration.

[0070] After testing, the final training set of the selected images, which converged and stabilized, had a total of 188,544 samples, including 151,644 pedestrian samples and 36,900 false positives. The test set had a total of 42,890 samples, including 19,580 pedestrian samples and 23,310 negative samples. The test set had 1,166 false positives and 764 false negatives. Therefore, the false positive rate was 2.72% and the false negative rate was 3.28%, both remaining at low levels.

[0071] The above process illustrates the training process of the vehicle blind spot pedestrian detection model. The following section introduces the application process of the vehicle blind spot pedestrian detection model.

[0072] Please refer to Figure 2 The diagram illustrates a flowchart of a blind spot pedestrian detection method according to an embodiment of this application. This method can also be applied to computer devices, which refer to electronic devices with data computing and processing capabilities, such as in-vehicle equipment. The method may include the following steps (210-220).

[0073] Step 210: Input the vehicle blind spot image in the first format into the blind spot pedestrian detection model to perform pedestrian detection and obtain the blind spot pedestrian detection result.

[0074] The blind spot pedestrian detection model is trained using the method described above. Specifically, it is an image processing model trained based on a first vehicle blind spot image set and the corresponding pedestrian annotation information. The pedestrian annotation information for the first vehicle blind spot image set is obtained by converting the pedestrian annotation information for a second vehicle blind spot image set. The second vehicle blind spot image set is obtained by converting the format of the first vehicle blind spot image set; the image format for the first vehicle blind spot image set is a first format, and the format for the second vehicle blind spot image set is a second format. The image data size for the first format is smaller than that for the second format.

[0075] Step 220: If the pedestrian detection result in the blind spot indicates that a pedestrian exists in the vehicle's blind spot image, output a prompt message.

[0076] When the vehicle is in motion, the monocular camera will capture a real-time three-channel YUV-NV12 (i.e., first format) image frame sequence of the blind spot while the vehicle is moving. The three-channel YUV-NV12 image frame sequence is input into the blind spot pedestrian detection model. If the model detects that there is a pedestrian in the blind spot corresponding to the current frame, it will issue a prompt to the driver, such as a voice prompt; if there is no pedestrian, it will continue to detect the next image frame.

[0077] In the technical solution provided in this application embodiment, the original vehicle blind spot image set in the first format is converted into the second format and then pedestrian annotation is performed. The annotation information of the second format image is then converted to obtain the annotation information under the first format image. The image processing model is trained directly using the vehicle blind spot image set in the first format with a smaller data volume and the converted pedestrian annotation information. This results in a blind spot pedestrian detection model with a smaller number of model parameters. Furthermore, the first format image detected by this model is relatively small. By using this model for blind spot pedestrian detection, the amount of data processing and the data processing time when the vehicle-mounted equipment performs pedestrian detection can be reduced, frame loss can be reduced, thereby reducing the false alarm rate and false negative rate of blind spot pedestrian detection and improving the accuracy of blind spot pedestrian warning.

[0078] Optionally, the pedestrian detection result in the blind spot includes the lateral distance between the pedestrian and the vehicle's blind spot. Accordingly, the onboard device can output different prompts based on the relationship between the lateral distance and a distance threshold, including but not limited to: If the lateral distance is less than or equal to the first distance threshold, output a level one alarm message; If the lateral distance is less than or equal to the second distance threshold, output a level 2 alarm message; If the lateral distance is less than or equal to the third distance threshold, output a level 3 alarm message.

[0079] The first distance threshold can be 3 meters, the second distance threshold can be 2 meters, and the third distance threshold can be 1 meter.

[0080] For example, the aforementioned blind spot pedestrian detection model, in addition to detecting the presence of pedestrians in the blind spot, can also detect whether the pedestrian is within the lateral danger distance range of the blind spot. If the model detects a pedestrian in the blind spot and that the pedestrian is within the lateral danger distance range, it issues an alarm, such as a voice prompt to the driver. If no pedestrian is present or the pedestrian is within the lateral safety distance range of the blind spot, it continues to detect the next image frame. This alarm can be divided into three levels, improving the alarm effectiveness.

[0081] The above explanations detail the training and application processes of the vehicle blind spot pedestrian detection model. The following section will further elaborate on these processes. Figure 3 The overall process is explained in order to demonstrate the inventive concept of the technical solution provided in this application. Figure 3 A flowchart of a dangerous driving behavior warning method based on a single-channel deep learning model is shown below: (1) Use a monocular camera to collect driving videos of the vehicle's blind spot while the vehicle is driving, and construct a sequence of YUV-NV12 format image frames; (2) Convert the sequence of YUV-NV12 format image frames into corresponding RGB images, obtain pedestrian annotation samples in the blind spot by annotating the RGB images, and convert the annotation samples into image coordinates in YUV-NV12 format; (3) Input the image coordinates and the YUV-NV12 format image set into the pedestrian detection model to obtain a deep learning detection model that can detect pedestrians in the blind spot; (4) Input the three-channel YUV-NV12 image into the blind spot pedestrian detection model to determine whether there are pedestrians in the blind spot image area. If the model detects that there are pedestrians in the blind spot, it will give a voice prompt to the driver. If there are no pedestrians, it will continue to detect the next frame image. The three-channel blind spot pedestrian detection model is generated based on the three-channel YUV-NV12 format blind spot image.

[0082] The following are embodiments of the apparatus of this application, which can be used to execute embodiments of the method of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method of this application.

[0083] Please refer to Figure 4 This diagram illustrates a block diagram of a training apparatus for a blind spot pedestrian detection model according to an embodiment of this application. The apparatus has the function of implementing the training method for the aforementioned blind spot pedestrian detection model; this function can be implemented in hardware or by hardware executing corresponding software. The apparatus 400 can be a computer device or can be installed within a computer device. The apparatus 400 may include: Image acquisition module 410 is used to acquire a first vehicle blind spot image set, wherein the image format corresponding to the first vehicle blind spot image set is a first format; The format conversion module 420 is used to convert the format of the images in the first vehicle blind spot image set to obtain a second vehicle blind spot image set. The image format corresponding to the second vehicle blind spot image set is a second format, and the image data volume corresponding to the first format is less than the image data volume corresponding to the second format. The annotation information acquisition module 430 is used to acquire the first pedestrian annotation information corresponding to the second vehicle blind spot image set; The annotation information conversion module 440 is used to convert the first pedestrian annotation information to obtain the second pedestrian annotation information corresponding to the first vehicle blind spot image set; The model training module 450 is used to train a preset image processing model based on the first vehicle blind spot image set and the second pedestrian annotation information to obtain a blind spot pedestrian detection model.

[0084] In the technical solution provided in this application embodiment, the original vehicle blind spot image set in the first format is converted into the second format and then pedestrian annotation is performed. The annotation information of the second format image is then converted to obtain the annotation information under the first format image. The image processing model is trained directly using the vehicle blind spot image set in the first format with a smaller data volume and the converted pedestrian annotation information. This results in a blind spot pedestrian detection model with a smaller number of model parameters. Furthermore, the first format image detected by this model is relatively small. Therefore, the data processing volume and data processing time when the blind spot pedestrian detection model performs pedestrian detection can be reduced, frame loss can be reduced, thereby reducing the false alarm rate and false negative rate of blind spot pedestrian detection and improving the accuracy of blind spot pedestrian warning.

[0085] Please refer to Figure 5 This diagram illustrates a block diagram of a blind spot pedestrian detection device according to an embodiment of this application. The device has the function of implementing the above-described blind spot pedestrian detection method; this function can be implemented in hardware or by hardware executing corresponding software. The device can be a computer device or can be installed within a computer device. The device 500 may include: The image detection module 510 is used to input the vehicle blind spot image in the first format into the blind spot pedestrian detection model for pedestrian detection, and obtain the blind spot pedestrian detection result; the blind spot pedestrian detection model is the blind spot pedestrian detection model trained by the above method; The information output module 520 is used to output a prompt message when the pedestrian detection result in the blind spot indicates that a pedestrian exists in the vehicle's blind spot image.

[0086] In the technical solution provided in this application embodiment, the original vehicle blind spot image set in the first format is converted into the second format and then pedestrian annotation is performed. The annotation information of the second format image is then converted to obtain the annotation information under the first format image. The image processing model is trained directly using the vehicle blind spot image set in the first format with a smaller data volume and the converted pedestrian annotation information. This results in a blind spot pedestrian detection model with a smaller number of model parameters. Furthermore, the first format image detected by this model is relatively small. By using this model for blind spot pedestrian detection, the amount of data processing and the data processing time when the vehicle-mounted equipment performs pedestrian detection can be reduced, frame loss can be reduced, thereby reducing the false alarm rate and false negative rate of blind spot pedestrian detection and improving the accuracy of blind spot pedestrian warning.

[0087] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0088] Please refer to Figure 6 This diagram illustrates a structural block diagram of a computer device according to an embodiment of this application. The computer device may be a terminal. This computer device is used to implement the training method or blind spot pedestrian detection method of the blind spot pedestrian detection model provided in the above embodiments. Specifically: Typically, computer device 900 includes a processor 901 and a memory 902.

[0089] Processor 901 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 901 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field Programmable Gate Array), and PLA (Programmable Logic Array). Processor 901 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 901 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 901 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0090] The memory 902 may include one or more computer-readable storage media, which may be non-transitory. The memory 902 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 902 is used to store at least one instruction, at least one program, code set, or instruction set, configured to be executed by one or more processors to implement the above-described blind spot pedestrian detection model training method or blind spot pedestrian detection method.

[0091] In some embodiments, the computer device 900 may optionally include a peripheral device interface 903 and at least one peripheral device. The processor 901, memory 902, and peripheral device interface 903 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 903 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency circuit 904, a touch display screen 905, a camera assembly 906, an audio circuit 907, a positioning assembly 908, and a power supply 909.

[0092] Those skilled in the art will understand that Figure 6The structure shown does not constitute a limitation on the computer device 900, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0093] In an exemplary embodiment, a computer-readable storage medium is also provided, the storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set, when executed by a processor, implements the above-described training method or blind spot pedestrian detection method for the blind spot pedestrian detection model.

[0094] Optionally, the computer-readable storage medium may include: ROM (Read Only Memory), RAM (Random Access Memory), SSD (Solid State Drives), or optical disc, etc. The random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0095] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the training method or blind spot pedestrian detection method of the aforementioned blind spot pedestrian detection model.

[0096] It should be understood that "multiple" as used herein refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. Furthermore, the step numbers described herein are merely illustrative of one possible execution order. In some other embodiments, the steps may not be executed in numerical order, such as two steps with different numbers being executed simultaneously, or two steps with different numbers being executed in the reverse order of the illustration. This application does not limit this.

[0097] In addition, in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0098] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A training method for a blind spot pedestrian detection model, characterized in that, The method includes: Acquire a first vehicle blind spot image set, wherein the image format corresponding to the first vehicle blind spot image set is a first format; The images in the first vehicle blind spot image set are converted to a second vehicle blind spot image set. The image format corresponding to the second vehicle blind spot image set is a second format. The image data volume corresponding to the first format is less than the image data volume corresponding to the second format. Obtain the first pedestrian annotation information corresponding to the second vehicle blind spot image set; The first pedestrian annotation information is converted to obtain the second pedestrian annotation information corresponding to the first vehicle blind spot image set; The image processing model is trained based on the first vehicle blind spot image set and the second pedestrian annotation information to obtain the blind spot pedestrian detection model.

2. The method according to claim 1, characterized in that, The first pedestrian labeling information includes first label information and second label information. The first label information is used to label pedestrians who are not obscured, and the second label information is used to label pedestrians who are obscured by objects.

3. The method according to claim 2, characterized in that, The obstruction includes at least one of a circular column or a rod-shaped object.

4. The method according to claim 1, characterized in that, The first pedestrian annotation information includes first pedestrian coordinate information; the step of converting the first pedestrian annotation information to obtain the second pedestrian annotation information corresponding to the first vehicle blind spot image set includes: The first pedestrian coordinate information is transformed to obtain the second pedestrian coordinate information corresponding to the first vehicle blind spot image set; Based on the label information corresponding to the first pedestrian coordinate information, determine the label information corresponding to the second pedestrian coordinate information; The second pedestrian annotation information is obtained based on the second pedestrian coordinate information and the corresponding label information.

5. The method according to claim 4, characterized in that, The method further includes: Acquire a third vehicle blind spot image set and pedestrian detection information corresponding to the third vehicle blind spot image set; The step of training a preset image processing model based on the first vehicle blind spot image set and the second pedestrian annotation information to obtain a blind spot pedestrian detection model includes: The first vehicle blind spot image set is filtered based on the second pedestrian annotation information, and the third vehicle blind spot image set is filtered based on the pedestrian detection information to obtain a filtered image training set. The filtered image training set includes vehicle blind spot images corresponding to at least two pedestrian feature conditions. The target image detection model is trained based on the selected image training set to obtain the blind spot pedestrian detection model.

6. The method according to claim 5, characterized in that, The step of filtering the first vehicle blind spot image set based on the second pedestrian annotation information and filtering the third vehicle blind spot image set based on the pedestrian detection information to obtain a filtered image training set includes: When the number of images in the first vehicle blind spot image set reaches a first quantity threshold, images in the first vehicle blind spot image set that meet the first pedestrian feature condition are selected to obtain the initial selected image training set. The preset image detection model is trained based on the initial selected image training set to obtain the intermediate blind spot pedestrian detection model. The pedestrian detection information is obtained by the intermediate blind spot pedestrian detection model detecting the third vehicle blind spot image set. Based on the pedestrian detection information, target images that meet the second pedestrian feature conditions in the third vehicle blind spot image set are added to the filtered image training set to obtain the updated filtered image training set. The step of training the target image detection model based on the selected image training set to obtain the blind spot pedestrian detection model includes: The intermediate blind spot pedestrian detection model is trained based on the updated selected image training set to obtain the blind spot pedestrian detection model.

7. The method according to claim 6, characterized in that, The acquisition of the third vehicle blind spot image set and the pedestrian detection information corresponding to the third vehicle blind spot image set includes: Obtain the blind spot image set of the third vehicle; Based on the aforementioned pedestrian detection model for the intermediate blind spot, pedestrian detection is performed on the images in the third vehicle blind spot image set to obtain the pedestrian detection information.

8. The method according to claim 7, characterized in that, After obtaining the pedestrian detection information by performing pedestrian detection on the images in the third vehicle blind spot image set based on the intermediate blind spot pedestrian detection model, the method further includes: Determine whether the pedestrian detection information meets the training constraints; If the pedestrian detection information does not meet the training constraints, determine the number of target images; If the number of target images is less than a preset threshold, the second pedestrian feature condition is updated, and the step of adding the target images that meet the second pedestrian feature condition in the third vehicle blind spot image set to the filtered image training set based on the pedestrian detection information is performed to obtain the updated filtered image training set.

9. The method according to claim 8, characterized in that, After determining whether the pedestrian detection information meets the training constraints, the method further includes: If the pedestrian detection information meets the training constraints, the current intermediate blind spot pedestrian detection model is determined as the blind spot pedestrian detection model.

10. The method according to any one of claims 5 to 9, characterized in that, The method further includes: Obtain anomaly detection video; Based on the blind spot pedestrian detection model, pedestrian detection is performed on the abnormal detection video to obtain an image with abnormal detection results; The selected image training set is updated based on the images with abnormal detection results; The blind spot pedestrian detection model is trained based on the updated selected image training set to obtain the updated blind spot pedestrian detection model.

11. The method according to any one of claims 1 to 9, characterized in that, The first format is YUV format, and the second format is RGB format.

12. A method for pedestrian detection in blind spots, characterized in that, The method includes: The vehicle blind spot image in the first format is input into the blind spot pedestrian detection model for pedestrian detection to obtain the blind spot pedestrian detection result; the blind spot pedestrian detection model is the blind spot pedestrian detection model trained by the method described in any one of claims 1 to 11; If the pedestrian detection result in the blind spot indicates that a pedestrian exists in the vehicle's blind spot image, a prompt message is output.

13. The method according to claim 12, characterized in that, The blind spot pedestrian detection result includes the lateral distance of the pedestrian to the vehicle's blind spot. When the blind spot pedestrian detection result indicates the presence of a pedestrian in the vehicle's blind spot image, a prompt message is output, including: If the lateral distance is less than or equal to the first distance threshold, output a level one alarm message; If the lateral distance is less than or equal to the second distance threshold, output a level 2 alarm message; If the lateral distance is less than or equal to the third distance threshold, a level three alarm message is output.

14. A training device for a blind spot pedestrian detection model, characterized in that, The device includes: The image acquisition module is used to acquire a first vehicle blind spot image set, wherein the image format corresponding to the first vehicle blind spot image set is a first format; The format conversion module is used to convert the format of the images in the first vehicle blind spot image set to obtain the second vehicle blind spot image set. The image format corresponding to the second vehicle blind spot image set is the second format, and the image data volume corresponding to the first format is less than the image data volume corresponding to the second format. The annotation information acquisition module is used to acquire the first pedestrian annotation information corresponding to the second vehicle blind spot image set; The annotation information conversion module is used to convert the first pedestrian annotation information to obtain the second pedestrian annotation information corresponding to the first vehicle blind spot image set; The model training module is used to train a preset image processing model based on the first vehicle blind spot image set and the second pedestrian annotation information to obtain a blind spot pedestrian detection model.

15. A pedestrian detection device for blind spots, characterized in that, The device includes: An image detection module is used to input a vehicle blind spot image in a first format into a blind spot pedestrian detection model for pedestrian detection, and obtain blind spot pedestrian detection results; the blind spot pedestrian detection model is a blind spot pedestrian detection model trained by the method described in any one of claims 1 to 11; The information output module is used to output a prompt message when the pedestrian detection result in the blind spot indicates that a pedestrian exists in the vehicle's blind spot image.

16. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one instruction, at least one program, a code set, or an instruction set, the at least one instruction, the at least one program, the code set, or the instruction set being loaded and executed by the processor to implement the training method for the blind spot pedestrian detection model as described in any one of claims 1 to 11, or the blind spot pedestrian detection method as described in claim 12 or 13.

17. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the training method for the blind spot pedestrian detection model as described in any one of claims 1 to 11, or the blind spot pedestrian detection method as described in claim 12 or 13.

Citation Information

Patent Citations

  • Method, system and device for detecting a riding vehicle in real time through vehicle-mounted look-around system

    CN111753593A

  • Non-motor vehicle illegal parking identification method and device, computer equipment and storage medium

    CN115082903A