Method for detecting behavior of using mobile phone by riding bicycle

Abstract

According to the method for detecting the behavior of using the mobile phone by riding, provided by the invention, the front driving image capable of being used for behavior analysis is selected from the images collected by the existing monitoring equipment, the images of the non-motor vehicle and the rider are intercepted and subjected to direction correction to form the video to be analyzed, and the video to be analyzed is used for detecting the behavior of using the mobile phone by riding. According to the method, recognizable behavior characteristics of a vehicle driver in the video are effectively improved, behavior recognition is performed through the three-dimensional convolutional neural network, and it is ensured that the behavior that a non-motor vehicle rider uses a mobile phone in the video can be effectively found; behavior recognition is carried out through the three-dimensional convolutional neural network, so that the error recognition rate can be effectively reduced; according to the method, hardware does not need to be newly added, and the mobile phone using behavior can be detected based on the video data with the common resolution ratio at low cost.

Description

Technical Field

[0001] The invention relates to the technical field of intelligent traffic control, and in particular to a method for detecting the behavior of using a mobile phone while riding a bicycle. Background Art

[0002] Common two-wheeled vehicles include bicycles, electric bicycles and motorcycles. The behavior of two-wheeled riders using mobile phones while riding is potentially very dangerous. The driver's attention is distracted and he is slow to react to the surrounding environment, which can easily lead to traffic accidents. At the same time, the harm of using mobile phones while riding is often difficult to detect. Among many cyclists, some people may have formed this bad habit, but have not yet caused serious consequences, thus ignoring its potential risks. Since mobile phone targets are relatively small and have fewer pixels in video images, existing target detection algorithms can only support the detection of mobile phone use while riding in some high-definition videos, and it is difficult to detect mobile phone use in videos with ordinary resolutions. Summary of the invention

[0003] In order to solve the problem that it is difficult to detect the use of mobile phones while riding a bicycle in videos with ordinary resolution, the present invention provides a method for detecting the use of mobile phones while riding a bicycle, which can effectively detect the use of mobile phones by non-motor vehicle riders in videos.

[0004] The technical solution of the present invention is as follows: a method for detecting the use of a mobile phone while riding a bicycle, characterized in that it comprises the following steps: S1: Build the target detection model and the frontal portrait recognition and classification model; Locating the image area where the non-motor vehicle is located in the image to be detected based on the target detection model; Based on the portrait recognition model, for the portrait part in the image, the portrait in the front direction is classified and recognized; S2: The entire image area is used as the area to be detected; S3: Obtain the image collected by the monitoring device, which is recorded as the image to be monitored; S4: based on the target detection model, locate the non-motor vehicle in the to-be-detected area in the to-be-detected image, cut out the image area where the non-motor vehicle and the driver are located, and record it as: target image area; for the portrait part in the target image area, identify whether it is a portrait in the front direction; If yes, it means that the non-motor vehicle captured in the image to be monitored is traveling in the front direction, which is recorded as the image to be analyzed; and step S4 is executed; Otherwise, the image to be monitored is discarded, and steps S2 to S3 are executed repeatedly until all the images to be monitored are processed; S5: Based on the image to be analyzed, continuously receive images collected by the monitoring device, find all images with the same vehicle and driver, and construct a set of images to be analyzed; S6: performing direction correction on the images of the non-motor vehicle and the cyclist in the target image area in each frame of the image set to be analyzed; S7: all corrected pictures in the set of pictures to be analyzed are collectively referred to as videos to be analyzed; S8: Building a mobile phone behavior detection model based on a three-dimensional convolutional neural network; The mobile phone use behavior detection model is used to determine whether the cyclist uses a mobile phone; The mobile phone use behaviors include: talking on the phone while riding a bicycle, playing with the phone with one hand, operating the phone on a stand with one hand, and playing with the phone with both hands; Building a training data set and a validation set based on historical data, training the mobile phone usage behavior detection model, and obtaining a trained mobile phone usage behavior detection model; S9: Input the video to be analyzed into the trained mobile phone usage behavior detection model to determine whether the cyclist uses the mobile phone.

[0005] It is further characterized by: The method for constructing the set of pictures to be analyzed comprises the following steps: a1: Initialize the image set to be analyzed; Put the image to be analyzed as the first frame image into the set of pictures to be analyzed; a2: constructing a rectangular frame of the target image area in the image to be analyzed, recorded as: the current image area rectangular frame; based on the position and size of the current image area rectangular frame, the current image area rectangular frame is enlarged to construct the area frame to be detected of the next frame of image; a3: Acquire the image to be detected of the next frame; a4: Based on the rectangular frame where the target image area is located, the side length is enlarged by a preset multiple N to establish a sliding scanning frame; Wherein, the area of ​​the sliding scanning frame is smaller than the area of ​​the area to be detected; a5: sliding the sliding scanning frame in the to-be-detected area frame of the to-be-detected image according to a preset step length to obtain M groups of image hash values ​​in the sliding scanning frames; a6: Calculate the hash value of the image inside the rectangular frame of the current image area in the image to be analyzed, which is recorded as: hash value for comparison; The hash values ​​of the images in the M groups of rectangular frames are respectively calculated with the comparison hash values ​​to obtain similarities, which are recorded as: hash value similarity; a7: Compare the hash value similarity of the M groups with a preset difference threshold; If all the hash value similarities are greater than the difference threshold, it means that there is no image similar to the image in the target image area in the M groups of sliding scanning frames; execute step a8; If there is a group of hash value similarities less than the difference threshold, it means that the image in the corresponding sliding scanning frame is most similar to the image in the target image area, which is recorded as: candidate image, and step a9 is executed; If there are multiple groups of hash value similarities less than the difference threshold, find the hash value similarity with the largest absolute value, and the image in the corresponding sliding scanning frame is most similar to the image in the target image area, and record it as the candidate image; execute step a9; a8: discard the image and execute steps a2 to a7 until the vehicle leaves the acquisition range of the monitoring device; a9: sort the candidate images by time and put them into the image set to be analyzed; a10: record the candidate image as the image to be analyzed, and execute steps a2 to a9 in a loop; In step a8, based on the preset driving-away threshold NP, if the images to be detected in consecutive NP frames are discarded, it is determined that the vehicle has left the acquisition range of the monitoring device; The method for constructing the to-be-detected area frame comprises the following steps: b1: Construct a coordinate system on the image to be monitored, with the upper left corner as the origin of the coordinate system; b2: taking the center of the current image area rectangular frame as the center of the to-be-detected area frame; b3: Get the side lengths of the current image area rectangle, recorded as: horizontal side and vertical side; Get the preset magnification CN, where CN>1; Taking CN times of the horizontal side as the horizontal side length of the to-be-detected area frame; Taking CN times of the vertical side as the vertical side length of the to-be-detected area frame; b4: in the next frame of image, constructing the to-be-detected area frame based on the center position of the to-be-detected area frame; In step S6, the direction correction method includes the following steps: c1: Obtain each frame of the picture in the picture set to be analyzed one by one, and record it as: the picture to be corrected; c2: Based on the direction of the road in the image to be corrected, the image of the road in the image to be corrected is corrected to the angle direction of the frontal shooting to obtain a correction matrix; c3: Based on the correction matrix, the images of the non-motor vehicle and the cyclist in the target image area in the picture are directional corrected to obtain a frontal perspective image; c4: storing the corrected images of the non-motor vehicle and the cyclist in the target image area to replace the original images, so as to obtain a corrected set of images to be analyzed; In step S7, all corrected pictures in the to-be-analyzed picture set are collectively referred to as a to-be-analyzed video, which specifically includes the following steps: d1: adjust all the images in the corrected image set to be analyzed to a fixed size of the specified size through image scaling; d2: Combine all the resized images into a video clip to obtain the video to be analyzed.

[0006] The present application provides a method for detecting the use of mobile phones while riding a bicycle. Based on images collected by existing monitoring equipment, a front driving image that can be used for behavior analysis is selected, and images of non-motor vehicles and cyclists are captured. After direction correction, a video to be analyzed is constructed, which effectively improves the recognizable behavior characteristics of vehicle drivers in the video, and then the behavior is recognized through a three-dimensional convolutional neural network to ensure that the use of mobile phones by non-motor vehicle cyclists in the video can be effectively discovered; the use of a three-dimensional convolutional neural network for behavior recognition can effectively reduce the misrecognition rate of cyclists' behavior; the method does not require additional hardware and can detect the use of mobile phones based on video data with ordinary resolution at a relatively low cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a flow chart of the method for detecting the use of mobile phones while riding a bicycle in this application; Figure 2 is an example of the area box to be detected; Figure 3 This is an example of direction correction. DETAILED DESCRIPTION

[0008] like Figure 1 As shown, the present invention includes a method for detecting the behavior of using a mobile phone while riding a bicycle, which includes the following steps.

[0009] S1: Build the target detection model and the frontal portrait recognition and classification model; Based on the target detection model, locate the image area where the non-motor vehicle is located in the image to be detected; Based on the portrait recognition model, the portrait part in the image is classified and recognized in the front direction.

[0010] In specific implementation, the target detection model is based on various non-motor vehicle recognition models in the prior art, such as: non-motor vehicle recognition models based on neural network CNN, YOLO, GAN and other models; the portrait recognition model is based on the machine learning classification model in the prior art, such as: support vector machine, decision tree or random forest, etc. A labeled training data set is constructed based on historical data to train the model, and then a validation set is constructed based on historical data to validate the model until a trained model is obtained.

[0011] S2: The entire image area is used as the area to be detected. When detecting the first image, the entire image range is used as the area to be monitored.

[0012] S3: Obtain the image collected by the monitoring device, recorded as: the image to be monitored.

[0013] In this method, the video data collected by the monitoring equipment with ordinary resolution is used as the data source, and the data is decomposed into image frames, and the image frames are recorded as the images to be monitored. The data of the image acquisition device installed based on the road test can also be used as the data source, and the pictures obtained by the image acquisition device are used as the images to be monitored. In this method, there is no need to change the resolution of the existing monitoring equipment, and the behavior detection can be achieved using the existing monitoring equipment.

[0014] S4: Based on the target detection model, locate the non-motor vehicle in the detection area of ​​the detection image, crop the image area where the non-motor vehicle and the driver are located, and record it as: the target image area; for the portrait part in the target image area, identify whether it is a frontal portrait based on the portrait recognition model; If yes, it means that the non-motor vehicle captured in the image to be monitored is traveling in the front direction, which is recorded as the image to be analyzed; and step S4 is executed; Otherwise, the image to be monitored is discarded, and steps S2 to S3 are executed in a loop until all the images to be monitored are processed.

[0015] The target image region construction method in this method is as follows: Construct a coordinate axis in the image to be detected, with the upper left corner as the origin. Figure 3 As shown. Find the point corresponding to the minimum x and y values ​​in the image pixel coordinates of the non-motor vehicle and the driver as the upper left corner, and use the point corresponding to the maximum x and y values ​​as the lower right corner to construct a rectangle, that is, to obtain the rectangular frame corresponding to the target image area.

[0016] In this method, the recognition area of ​​the portrait recognition model is set in the target image area rather than the area to be detected, which can effectively reduce the amount of calculation and improve the calculation efficiency.

[0017] S5: Continuously receive the images collected by the monitoring device based on the images to be analyzed, find all the images with the same vehicle and driver therein, and construct a set of images to be analyzed.

[0018] A method for constructing a set of images to be analyzed includes the following steps.

[0019] a1: Initialize the set of images to be analyzed; Take the image to be analyzed as the first-frame image and put it into the set of images to be analyzed.

[0020] a2: Construct a rectangular frame for the target image area in the image to be analyzed, denoted as: the rectangular frame of the current image area; based on the position and size of the rectangular frame of the current image area, after enlarging the rectangular frame of the current image area, construct a detection area frame for the next-frame image; a3: Obtain the image to be detected for the next frame.

[0021] a4: Based on the rectangular frame where the target image area is located, enlarge the side length by a preset multiple N to establish a sliding scanning frame; among them, because it is necessary to meet the condition: the area of the sliding scanning frame is smaller than the area of the detection area. In this actual example, the detection area frame is enlarged by 1.2 (CN) times based on the side length of the rectangular frame where the target image area is located, then set the value of N as: 0.9 < N < CN; the specific value is set according to the calculation accuracy requirements. If the value of N is set small, the calculation amount is large and the accuracy is high. If the value of N is set large, the calculation amount is small and the accuracy is low. In this embodiment, the value of N is 0.9 to 1.1 times.

[0022] a5: In the detection area frame of the image to be detected, slide the sliding scanning frame at a preset step length to obtain the image hash values within M groups of sliding scanning frames.

[0023] The image hash value is to convert an image into a unique numerical identifier through a specific algorithm, and is often used for image similarity comparison. The image hash algorithm converts an image into a hash value of a fixed length and measures the similarity of images by comparing the Hamming distance of these hash values. The specific hash value calculation method and hash value similarity algorithm are implemented based on the existing technology.

[0024] a6: Calculate the hash value of the image inside the rectangular frame of the current image area in the image to be analyzed, denoted as: the comparison hash value; Calculate the similarity between the image hash values within M groups of rectangular frames and the comparison hash value respectively, denoted as: the hash value similarity.

[0025] a7: Compare the hash value similarities of M groups with a preset difference threshold; In this application, the similarity accuracy of two images compared with each other is controlled based on the difference threshold, and the specific value is set according to the actual accuracy requirements.

[0026] If all hash value similarities are greater than the difference threshold, it means that there is no image similar to the image in the target image area in the M group of sliding scanning frames; execute step a8; If there is a set of hash value similarities less than the difference threshold, it means that the image in the corresponding sliding scanning frame is most similar to the image in the target image area, which is recorded as: candidate image, and step a9 is executed; If there are multiple groups of hash value similarities less than the difference threshold, find the hash value similarity with the largest absolute value, and the image in the corresponding sliding scanning box is most similar to the image in the target image area, which is recorded as: candidate image; execute step a9.

[0027] a8: Discard the image and execute steps a2 to a7 until the vehicle leaves the acquisition range of the monitoring device.

[0028] a9: Sort the candidate images by time and put them into the image set to be analyzed.

[0029] a10: record the candidate image as the image to be analyzed, and execute steps a2 to a9 in a loop.

[0030] In this application, through the process of constructing a set of pictures to be analyzed, video frames in normal resolution videos that cannot be used for recognition due to low resolution or because the driver is not in a frontal image and cannot be used for detection are removed, and all sufficiently clear pictures are retained based on hash value comparison, effectively improving the accuracy of subsequent three-dimensional convolutional neural network for behavior recognition.

[0031] In step a8, based on the preset departure threshold NP, if consecutive NP frames of images to be detected are discarded, it is determined that the vehicle has left the acquisition range of the monitoring device; in this embodiment, the NP value is 5, that is, if there is no target in 5 consecutive frames of images to be detected, it is determined that the vehicle has left.

[0032] like Figure 2 As shown, the method for constructing the to-be-detected area frame includes the following steps: b1: Construct a coordinate system on the image to be monitored, with the upper left corner as the origin of the coordinate system.

[0033] b2: Take the center of the current image area rectangle as the center of the area to be detected.

[0034] b3: Get the side lengths of the current image area rectangle, recorded as: horizontal side and vertical side; Get a preset magnification CN, where CN>1; in this method, the amount of calculation is reduced by limiting the target detection range in the area to be detected, and by setting CN to a value greater than 1, the area to be detected frame is gradually enlarged based on the area where the non-motor vehicle image in the previous frame image is located, so as to adapt to the increasing changes in the image area where the non-motor vehicle is located in the forward-moving video, ensuring that this application is more practical. In this embodiment, the value of CN is 1.2.

[0035] The horizontal length of the box of the area to be detected is CN times the horizontal length. The vertical side length of the to-be-detected area frame is CN times the vertical side length.

[0036] b4: In the next frame of the image, a frame of the area to be detected is constructed based on the center position of the frame of the area to be detected.

[0037] S6: Based on the fact that “the vehicle is traveling along the road”, the images of the non-motor vehicle and the cyclist in the target image area in each frame of the image set to be analyzed are directional corrected according to the traveling direction of the non-motor vehicle.

[0038] In step S6, the direction correction method includes the following steps: c1: Get each frame of the image set to be analyzed one by one, and record it as the image to be corrected; c2: Based on the direction of the road in the image to be corrected, the image of the road in the image to be corrected is corrected to the angle direction of the frontal shooting to obtain a correction matrix; c3: Based on the correction matrix, the images of non-motor vehicles and cyclists in the target image area in the picture are oriented to obtain a frontal perspective image; c4: The corrected images of non-motor vehicles and cyclists in the target image area are stored to replace the original images, and a corrected set of images to be analyzed is obtained.

[0039] like Figure 3 As shown, in the front-view image, the road direction r is perpendicular to the x-axis direction from top to bottom.

[0040] The direction r' of the road in the image to be corrected is at an angle α to the direction r. The direction r' of the road in the image to be corrected is first corrected to the direction r, and then the correction matrix corresponding to the image to be corrected is obtained. Then, based on this correction matrix, the images of non-motor vehicles and cyclists in the target image area in the image to be corrected are corrected to obtain the corrected frontal perspective image. The angle α can be obtained according to the parameters when the image acquisition device is installed, or it can be calculated based on the existing image recognition technology.

[0041] S7: All corrected pictures in the to-be-analyzed picture set are collectively referred to as a to-be-analyzed video.

[0042] In step S7, all corrected pictures in the to-be-analyzed picture set are collectively referred to as a to-be-analyzed video, which specifically includes the following steps: d1: All the pictures in the corrected picture set to be analyzed are adjusted to a fixed size of a specified size by image scaling; in this embodiment, the images are adjusted to a fixed size of 320*320 by image scaling.

[0043] d2: Combine all the resized images into a video clip to obtain the video to be analyzed.

[0044] The image frame values ​​in the video to be analyzed in this method include the rectified images of non-motor vehicles and cyclists, which can more effectively provide features for detection model recognition. Moreover, compared with the detection method based on the entire image, it effectively reduces the amount of calculation and improves the calculation efficiency.

[0045] S8: Building a mobile phone behavior detection model based on a three-dimensional convolutional neural network; The mobile phone behavior detection model is used to determine whether the cyclist uses a mobile phone; Mobile phone use behaviors include: talking on the phone while riding a bicycle, playing with the phone with one hand, operating the phone on a stand with one hand, and playing with the phone with both hands; A training data set and a validation set are constructed based on historical data, and the mobile phone usage behavior detection model is trained to obtain a trained mobile phone usage behavior detection model.

[0046] S9: Input the video to be analyzed into the trained mobile phone usage behavior detection model to determine whether the cyclist uses the mobile phone.

[0047] After using the technical solution of the present invention, based on existing equipment, four types of behaviors of using a mobile phone while riding a bicycle, namely, making phone calls while riding a bicycle, playing with a mobile phone with one hand, operating a mobile phone on a bracket with one hand, and playing with a mobile phone with both hands, can be effectively detected from videos of ordinary resolution.

Claims

1. A method for detecting the use of mobile phones while riding a bicycle, characterized in that: It includes the following steps: S1: Build the target detection model and the frontal portrait recognition and classification model; Locating the image area where the non-motor vehicle is located in the image to be detected based on the target detection model; Based on the portrait recognition model, for the portrait part in the image, the portrait in the front direction is classified and recognized; S2: The entire image area is used as the area to be detected; S3: Obtain the image collected by the monitoring device, which is recorded as the image to be monitored; S4: based on the target detection model, locate the non-motor vehicle in the to-be-detected area in the to-be-detected image, cut out the image area where the non-motor vehicle and the driver are located, and record it as: target image area; for the portrait part in the target image area, identify whether it is a portrait in the front direction; If yes, it means that the non-motor vehicle captured in the image to be monitored is traveling in the front direction, which is recorded as the image to be analyzed; and step S4 is executed; Otherwise, the image to be monitored is discarded, and steps S2 to S3 are executed repeatedly until all the images to be monitored are processed; S5: Based on the image to be analyzed, continuously receive images collected by the monitoring device, find all images with the same vehicle and driver, and construct a set of images to be analyzed; S6: performing direction correction on the images of the non-motor vehicle and the cyclist in the target image area in each frame of the image set to be analyzed; S7: all corrected pictures in the set of pictures to be analyzed are collectively referred to as videos to be analyzed; S8: Building a mobile phone behavior detection model based on a three-dimensional convolutional neural network; The mobile phone use behavior detection model is used to determine whether the cyclist uses a mobile phone; The mobile phone use behaviors include: talking on the phone while riding a bicycle, playing with the phone with one hand, operating the phone on a stand with one hand, and playing with the phone with both hands; Building a training data set and a validation set based on historical data, training the mobile phone usage behavior detection model, and obtaining a trained mobile phone usage behavior detection model; S9: Input the video to be analyzed into the trained mobile phone usage behavior detection model to determine whether the cyclist uses the mobile phone.

2. A method for detecting the use of a mobile phone while riding a bicycle according to claim 1, characterized in that: The method for constructing the set of pictures to be analyzed comprises the following steps: a1: Initialize the image set to be analyzed; Put the image to be analyzed as the first frame image into the set of pictures to be analyzed; a2: constructing a rectangular frame of the target image area in the image to be analyzed, recorded as: the current image area rectangular frame; based on the position and size of the current image area rectangular frame, the current image area rectangular frame is enlarged to construct the area frame to be detected of the next frame of image; a3: Acquire the image to be detected of the next frame; a4: Based on the rectangular frame where the target image area is located, the side length is enlarged by a preset multiple N to establish a sliding scanning frame; Wherein, the area of ​​the sliding scanning frame is smaller than the area of ​​the area to be detected; a5: sliding the sliding scanning frame in the to-be-detected area frame of the to-be-detected image according to a preset step length to obtain M groups of image hash values ​​in the sliding scanning frames; a6: Calculate the hash value of the image inside the rectangular frame of the current image area in the image to be analyzed, which is recorded as: hash value for comparison; The hash values ​​of the images in the M groups of rectangular frames are respectively calculated with the comparison hash values ​​to obtain similarities, which are recorded as: hash value similarity; a7: Compare the hash value similarity of the M groups with a preset difference threshold; If all the hash value similarities are greater than the difference threshold, it means that there is no image similar to the image in the target image area in the M groups of sliding scanning frames; execute step a8; If there is a group of hash value similarities less than the difference threshold, it means that the image in the corresponding sliding scanning frame is most similar to the image in the target image area, which is recorded as: candidate image, and step a9 is executed; If there are multiple groups of hash value similarities less than the difference threshold, find the hash value similarity with the largest absolute value, and the image in the corresponding sliding scanning frame is most similar to the image in the target image area, and record it as the candidate image; execute step a9; a8: discard the image and execute steps a2 to a7 until the vehicle leaves the acquisition range of the monitoring device; a9: sort the candidate images by time and put them into the image set to be analyzed; a10: record the candidate image as the image to be analyzed, and execute steps a2 to a9 in a loop.

3. A method for detecting the use of a mobile phone while riding a bicycle according to claim 2, characterized in that: In step a8, based on the preset departure threshold NP, if the images to be detected in consecutive NP frames are discarded, it is determined that the vehicle has left the acquisition range of the monitoring device.

4. A method for detecting the use of a mobile phone while riding a bicycle according to claim 2, characterized in that: The method for constructing the to-be-detected area frame comprises the following steps: b1: Construct a coordinate system on the image to be monitored, with the upper left corner as the origin of the coordinate system; b2: taking the center of the current image area rectangular frame as the center of the to-be-detected area frame; b3: Get the side lengths of the current image area rectangle, recorded as: horizontal side and vertical side; Get the preset magnification CN, where CN>1; Taking CN times of the horizontal side as the horizontal side length of the to-be-detected area frame; Taking CN times of the vertical side as the vertical side length of the to-be-detected area frame; b4: In the next frame of image, the to-be-detected area frame is constructed based on the center position of the to-be-detected area frame.

5. According to claim 1, a method for detecting the use of mobile phones while riding a bicycle is characterized by: In step S6, the direction correction method includes the following steps: c1: Obtain each frame of the picture in the picture set to be analyzed one by one, and record it as: the picture to be corrected; c2: Based on the direction of the road in the image to be corrected, the image of the road in the image to be corrected is corrected to the angle direction of the frontal shooting to obtain a correction matrix; c3: Based on the correction matrix, the images of the non-motor vehicle and the cyclist in the target image area in the picture are directional corrected to obtain a frontal perspective image; c4: The corrected images of non-motor vehicles and cyclists in the target image area are stored to replace the original images, so as to obtain a corrected set of images to be analyzed.

6. The method for detecting the use of mobile phones while riding a bicycle according to claim 1, characterized in that: In step S7, all corrected pictures in the to-be-analyzed picture set are collectively referred to as a to-be-analyzed video, which specifically includes the following steps: d1: adjust all the images in the corrected image set to be analyzed to a fixed size of the specified size through image scaling; d2: Combine all the resized images into a video clip to obtain the video to be analyzed.

Images