Video switching identification method, device, equipment and medium

Abstract

The present invention discloses a method, apparatus, device, and medium for identifying video switching. The method comprises: obtaining a first feature map corresponding to each video frame in a selected video frame sequence based on a first feature extraction weight, wherein the first feature extraction weight decreases sequentially based on the frame order of the selected video frame sequence; obtaining a second feature map corresponding to each video frame in the selected video frame sequence based on a second feature extraction weight, wherein the second feature extraction weight increases sequentially based on the frame order of the selected video frame sequence; and determining whether an intermediate frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map. The present invention improves the efficiency of video segmentation and recognition.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a video switching recognition method, device, equipment and medium. Background Art

[0002] In recent years, short video applications have seen explosive growth across major internet platforms. These videos are typically produced through video editing, which involves splitting and merging video sources, then re-encoding them to generate new videos with varying expressiveness. Traditionally, video segmentation and recognition is performed manually with the help of software. This process not only requires manual intervention but is also slow and inefficient for large amounts of video. Summary of the Invention

[0003] The main purpose of the present invention is to provide a video switching recognition method, device, equipment and medium, aiming to solve the problem of low efficiency in video splitting and recognition.

[0004] To achieve the above object, the present invention provides a method for identifying video switching, which includes the following steps:

[0005] Obtaining a first feature map corresponding to each video frame in the selected video frame sequence according to a first feature extraction weight, wherein the first feature extraction weight decreases in sequence based on a frame order of the selected video frame sequence;

[0006] Obtaining a second feature map corresponding to each video frame in the selected video frame sequence according to a second feature extraction weight, wherein the second feature extraction weight increases sequentially based on a frame order of the selected video frame sequence;

[0007] According to the first feature map and the second feature map, it is determined whether the middle frame in the selected video frame sequence is a switching video frame.

[0008] In one embodiment, the first feature map corresponding to each of the video frames is determined by a first grouped convolutional network, and the second feature map corresponding to each of the video frames is determined by a second grouped convolutional network; the first grouped convolutional network and the second grouped convolutional network are both provided with convolution groups corresponding one-to-one to the video frames.

[0009] In one embodiment, in the first grouped convolutional network, the convolution group corresponding to each of the video frames is associated with the first feature extraction weight, and the first feature extraction weight decreases successively with the frame order of the selected video frame sequence; in the second grouped convolutional network, the convolution group corresponding to each of the video frames is associated with the second feature extraction weight, and the second feature extraction weight increases successively with the frame order of the selected video frame sequence.

[0010] In one embodiment, before the step of obtaining the first feature map corresponding to each video frame in the selected video frame sequence according to the first feature extraction weight, the method further includes:

[0011] Get multiple video frames of the video to be processed;

[0012] Determine a starting frame, a window length, and a sliding step length in the multi-frame video frame, wherein the window length is an odd value;

[0013] The selected video frame sequence is acquired according to the starting frame, the window length, and the sliding step size.

[0014] In one embodiment, the step of determining whether the middle frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map includes:

[0015] fusing the first feature map and the second feature map;

[0016] The fusion result of the first feature map and the second feature map is input into a preset two-dimensional convolutional network to obtain a recognition result of whether the intermediate frame is a switching video frame.

[0017] In one embodiment, the step of determining whether the middle frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map includes:

[0018] comparing the similarity between the first feature map and the second feature map;

[0019] When the similarity is less than or equal to a preset threshold, determining that the middle frame in the selected video frame sequence is a non-switching video frame;

[0020] When the similarity is greater than a preset threshold, the middle frame in the selected video frame sequence is determined to be a switching video frame.

[0021] In one embodiment, after the step of determining whether the middle frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map, the method further includes:

[0022] Obtaining recognition results of each video frame corresponding to the video to be processed;

[0023] The continuous video frames that are non-switched video frames as the recognition results are determined as split videos, and the split videos are continuously shot by the same lens.

[0024] To achieve the above-mentioned object, the present invention further provides a video switching recognition device, the video switching recognition device comprising:

[0025] A first determining module is configured to obtain a first feature map corresponding to each video frame in a selected video frame sequence according to a first feature extraction weight, wherein the first feature extraction weight decreases in sequence based on a frame order of the selected video frame sequence;

[0026] a second determining module, configured to obtain a second feature map corresponding to each video frame in the selected video frame sequence according to a second feature extraction weight, wherein the second feature extraction weight increases sequentially based on a frame order of the selected video frame sequence;

[0027] An identification module is used to determine whether an intermediate frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map.

[0028] To achieve the above-mentioned purpose, the present invention also provides a video switching recognition device, which includes a memory, a processor, and a video switching recognition program stored in the memory and executable on the processor. When the video switching recognition program is executed by the processor, it implements the various steps of the video switching recognition method described above.

[0029] To achieve the above objectives, the present invention also provides a computer-readable storage medium, which stores a video switching recognition program. When the video switching recognition program is executed by a processor, it implements the various steps of the video switching recognition method described above.

[0030] The present invention provides a method, apparatus, device, and medium for identifying video switching. The method comprises obtaining a first feature map corresponding to each video frame in a selected video frame sequence based on a first feature extraction weight, wherein the first feature extraction weight decreases sequentially based on the frame order of the selected video frame sequence; obtaining a second feature map corresponding to each video frame in the selected video frame sequence based on a second feature extraction weight, wherein the second feature extraction weight increases sequentially based on the frame order of the selected video frame sequence; and determining whether an intermediate frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map. Because the first feature map focuses on extracting features of video frames ranked higher, and the second feature map focuses on extracting features of video frames ranked lower, the recognition result of the intermediate frame is comprehensively determined based on the first and second feature maps, thereby improving the accuracy and efficiency of identifying switching frames in the video and enhancing the efficiency of video segmentation and recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of the hardware structure of a video switching recognition device according to an embodiment of the present invention;

[0032] Figure 2 1 is a flow chart of a first embodiment of a method for identifying video switching according to the present invention;

[0033] Figure 3 Schematic diagram of N groups of convolutional layers in the video switching recognition method of the present invention;

[0034] Figure 4 Schematic diagram of a group convolutional network in the video switching recognition method of the present invention;

[0035] Figure 5 A schematic diagram of switching video frames in the video switching identification method of the present invention;

[0036] Figure 6 A schematic diagram of switching video frames in the video switching identification method of the present invention;

[0037] Figure 7 A schematic diagram of switching video frames in the video switching identification method of the present invention;

[0038] Figure 8 4 is a detailed flowchart of step S30 of the second embodiment of the video switching identification method of the present invention;

[0039] Figure 9 Schematic diagram of the logical structure of the video switching identification device of the present invention.

[0040] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0041] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] The main solution of an embodiment of the present invention is: obtaining a first feature map corresponding to each video frame in a selected video frame sequence according to a first feature extraction weight, wherein the first feature extraction weight decreases in sequence based on the frame sequence of the selected video frame sequence; obtaining a second feature map corresponding to each video frame in the selected video frame sequence according to a second feature extraction weight, wherein the second feature extraction weight increases in sequence based on the frame sequence of the selected video frame sequence; determining whether the intermediate frame in the selected video frame sequence is a switching video frame according to the first feature map and the second feature map.

[0043] Since the first feature map focuses on extracting the features of the video frames with higher rankings, and the second feature map focuses on extracting the features of the video frames with lower rankings, the recognition results of the intermediate frames are comprehensively determined based on the first feature map and the second feature map, which improves the accuracy and efficiency of the recognition of switching frames in the video and improves the efficiency of video splitting and recognition.

[0044] As an implementation solution, the video switching recognition device can be as follows Figure 1 shown.

[0045] The embodiment of the present invention relates to a video switching recognition device, which includes a processor 101, such as a CPU, a memory 102, and a communication bus 103. The communication bus 103 is used to implement connection and communication between these components.

[0046] The memory 102 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. Figure 1 As shown, the memory 102 as a computer-readable storage medium may include a video switching recognition program; and the processor 101 may be used to call the video switching recognition program stored in the memory 102 and perform the following operations:

[0047] Obtaining a first feature map corresponding to each video frame in the selected video frame sequence according to a first feature extraction weight, wherein the first feature extraction weight decreases in sequence based on a frame order of the selected video frame sequence;

[0048] Obtaining a second feature map corresponding to each video frame in the selected video frame sequence according to a second feature extraction weight, wherein the second feature extraction weight increases sequentially based on a frame order of the selected video frame sequence;

[0049] According to the first feature map and the second feature map, it is determined whether the middle frame in the selected video frame sequence is a switching video frame.

[0050] In one embodiment, the processor 101 may be configured to call a video switching recognition program stored in the memory 102 and perform the following operations:

[0051] Get multiple video frames of the video to be processed;

[0052] Determine a starting frame, a window length, and a sliding step length in the multi-frame video frame, wherein the window length is an odd value;

[0053] The selected video frame sequence is acquired according to the starting frame, the window length, and the sliding step size.

[0054] In one embodiment, the processor 101 may be configured to call a video switching recognition program stored in the memory 102 and perform the following operations:

[0055] fusing the first feature map and the second feature map;

[0056] The fusion result of the first feature map and the second feature map is input into a preset two-dimensional convolutional network to obtain a recognition result of whether the intermediate frame is a switching video frame.

[0057] In one embodiment, the processor 101 may be configured to call a video switching recognition program stored in the memory 102 and perform the following operations:

[0058] comparing the similarity between the first feature map and the second feature map;

[0059] When the similarity is less than or equal to a preset threshold, determining that the middle frame in the selected video frame sequence is a non-switching video frame;

[0060] When the similarity is greater than a preset threshold, the middle frame in the selected video frame sequence is determined to be a switching video frame.

[0061] In one embodiment, the processor 101 may be configured to call a video switching recognition program stored in the memory 102 and perform the following operations:

[0062] Obtaining recognition results of each video frame corresponding to the video to be processed;

[0063] The continuous video frames that are non-switched video frames as the recognition results are determined as split videos, and the split videos are continuously shot by the same lens.

[0064] Based on the hardware architecture of the above-mentioned video switching recognition device, an embodiment of the video switching recognition method of the present invention is proposed.

[0065] Reference Figure 2 , Figure 2 This is a first embodiment of a video switching recognition method of the present invention, and the video switching recognition method includes the following steps:

[0066] Step S10 , obtaining a first feature map corresponding to each video frame in the selected video frame sequence according to a first feature extraction weight, wherein the first feature extraction weight decreases in sequence based on a frame order of the selected video frame sequence.

[0067] Specifically, the selected video frame sequence is a plurality of continuous video frames of the video to be processed, and the number of frames of the selected video frame sequence is an odd number, for example, the number of frames of the selected video frame sequence is 5. Feature extraction of the selected video frame sequence is more inclined to features before the middle frame of the selected video frame sequence, and the first feature extraction weight of the first feature graph decreases in sequence based on the frame order of the selected video frame sequence. The first feature extraction weight corresponding to each video frame is a fixed value. For example, the first feature extraction weight of the first frame of the selected video frame sequence is 1, the first feature extraction weight of the second frame of the video frame is 0.8, the first feature extraction weight of the third frame of the video frame is 0.6, the first feature extraction weight of the fourth frame of the video frame is 0.4, and the first feature extraction weight of the fifth frame of the video frame is 0.2.

[0068] Optionally, the first feature map corresponding to each video frame is determined by a first grouped convolutional network, and the first grouped convolutional network is provided with a convolution group corresponding to each video frame. Figure 3 As shown, the first convolutional network includes N groups of convolution groups, each of which is a 1x1 grouped convolution. The grouped convolutional network is as follows: Figure 4 As shown, the video frames are input into the corresponding convolution groups for feature extraction to obtain a first feature map. Optionally, the first group convolution network also includes a two-dimensional convolutional neural network, and the feature maps output by the multiple convolution groups are input into the two-dimensional convolutional neural network for further feature extraction to obtain the first feature map.

[0069] Optionally, in the first grouped convolutional network, the convolution group corresponding to each video frame is associated with a first feature extraction weight, and the first feature extraction weight decreases successively with the frame order of the selected video frame sequence, wherein the first feature extraction weights of the N groups of convolution groups are fixed weights, and the first feature extraction weights of each group of convolution groups are w1=1.0-n / N, n=0~(N-1), N is an odd number. Exemplarily, the first grouped convolutional network includes 5 groups of convolution groups, the first feature extraction weight of the first convolution group is 1, the first feature extraction weight of the second convolution group is 0.8, the first feature extraction weight of the third convolution group is 0.6, the first feature extraction weight of the fourth convolution group is 0.4, and the first feature extraction weight of the fifth convolution group is 0.2.

[0070] Prior to step S10, multiple video frames of the video to be processed are obtained; the starting frame, window length, and sliding step size of the multiple video frames are determined, where the window length is an odd value; and a selected video frame sequence is obtained based on the starting frame, window length, and sliding step size. For example, the video to be processed is M frames, the selected video frame sequence is N frames, for example, N is 5, the starting frame is the first frame of the video to be processed, the window length is 5, the selected video frame sequence is frames 1-5, the sliding step size is 1, the window slides to the second frame, the window length is 5, and the selected video frame sequence is frames 2-6, until the selected video frame sequence is determined to be M-4 frames to M frames.

[0071] Before the selected video frame sequence is input into the first grouped convolutional network, for N consecutive video frames Nx(C,H,W), the channel dimension is merged (NxC,H,W) to obtain a multi-channel image, and the multi-channel image is input into the first grouped convolutional network. Wherein, N represents N frames of video frames, C represents the image channel, which is generally 3, for example, BGR means 3 channels of blue, green and red, H represents the height of the video frame, and W represents the width of the video frame. For example, 5 consecutive frames of images 5*(3,1920,1080), with a resolution of 1920x1080 and 3 channels of blue, green and red, are 5 independent video frames before merging, and a 15-channel image (15,1920,1080) is formed after merging.

[0072] When a selected video frame sequence is input into the first grouped convolutional network, the multi-channel image (N*C,H,W) is automatically split into N data of dimensions (C,H,W) along the channel dimension. These data are then fed into the N convolution groups in the first grouped convolutional network. Each convolution group outputs a feature map containing C channels of dimensions (C,H,W). Optionally, the feature map output by each convolution group is fed into a two-dimensional convolutional network to produce the first feature map.

[0073] Step S20: obtaining a second feature map corresponding to each video frame in the selected video frame sequence according to a second feature extraction weight, wherein the second feature extraction weight is increased sequentially based on a frame order of the selected video frame sequence.

[0074] Specifically, the feature extraction of the selected video frame sequence is more inclined to the features after the middle frame of the selected video frame sequence, and the second feature extraction weight of the second feature map increases sequentially based on the frame order of the selected video frame sequence. For example, the second feature extraction weight of the first frame of the video frame in the selected video frame sequence is 0.2, the second feature extraction weight of the second frame of the video frame is 0.4, the second feature extraction weight of the third frame of the video frame is 0.6, the second feature extraction weight of the fourth frame of the video frame is 0.8, and the second feature extraction weight of the fifth frame of the video frame is 1.

[0075] Optionally, a second feature map corresponding to each video frame is determined by a second grouped convolutional network; the second grouped convolutional network is provided with a convolution group corresponding to each video frame. Figure 3 As shown, the first convolutional network includes N groups of convolution groups, each of which is a 1x1 grouped convolution. The grouped convolutional network is as follows: Figure 4 As shown, the video frames are input into the corresponding convolution groups for feature extraction to obtain a second feature map. Optionally, the second grouped convolution network further includes a two-dimensional convolutional neural network, and the feature maps output by the multiple convolution groups are input into the two-dimensional convolutional neural network for further feature extraction to obtain a second feature map.

[0076] Optionally, in the second grouped convolutional network, the convolution group corresponding to each video frame is associated with a second feature extraction weight, and the second feature extraction weight increases sequentially with the frame order of the selected video frame sequence, wherein the second feature extraction weights of the N groups of convolution groups are fixed weights, and the second feature extraction weight of each group of convolution groups is w2=n / N, where n=1~N, and N is an odd number. Exemplarily, the second grouped convolutional network includes 5 convolution groups, the second feature extraction weight of the first convolution group is 0.2, the second feature extraction weight of the second convolution group is 0.4, the second feature extraction weight of the third convolution group is 0.6, the second feature extraction weight of the fourth convolution group is 0.8, and the second feature extraction weight of the fifth convolution group is 1.

[0077] Before the selected video frame sequence is input into the second grouped convolutional network, for N consecutive video frames Nx(C,H,W), the channel dimension is merged (NxC,H,W) to obtain a multi-channel image, and the multi-channel image is input into the second grouped convolutional network. Wherein, N represents N frames of video frames, C represents the image channel, which is generally 3, for example, BGR means 3 channels of blue, green and red, H represents the height of the video frame, and W represents the width of the video frame. For example, 5 consecutive frames of images 5*(3,1920,1080), 1920x1080 is the resolution, and the channels are 3 channels of blue, green and red. Before merging, there are 5 independent video frames, and after merging, a 15-channel image (15,1920,1080) is formed.

[0078] When a selected video frame sequence is fed into the second grouped convolutional network, the multi-channel image (N*C,H,W) is automatically split into N data sets of (C,H,W) along the channel dimension. These are then fed into the N convolution groups in the second grouped convolutional network. Each convolution group outputs a feature map containing C channels of (C,H,W) dimensions. Optionally, the feature map output by each convolution group is fed into a two-dimensional convolutional network to produce a second feature map.

[0079] Step S30: Determine whether the middle frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map.

[0080] Specifically, based on the first feature map and the second feature map, it is determined whether the middle frame in the selected video frame sequence is a switching video frame, wherein the selected video frame sequence has N video frames, the middle frame in the selected video frame is 1+(N-1) / 2 frames, and the first (N-1) / 2 video frames and the last (N-1) / 2 video frames are used to identify whether the middle frame (the (N+1) / 2th frame) is a switching frame. The middle frame is a non-switching video frame, which means that the middle frame is a picture shot by the same camera lens. The middle frame is a switching video frame, which means that the middle frame is a picture shot by one camera switched to a picture shot by another camera. For example, Figure 5 As shown, the three video frames are all switching video frames, such as Figure 6 As shown in FIG, 6 video frames are all switching video frames. The switching video frames can be the conversion frames of the video frames shot by two lenses, such as Figure 7 As shown, there is no transition frame between the two video frames, and the two video frames are both switching video frames.

[0081] Optionally, based on the first feature map and the second feature map, determine whether the middle frame in the selected video frame sequence is a switching video frame, and compare the similarity between the first feature map and the second feature map; when the similarity is less than or equal to a preset threshold, it means that the video pictures of the first feature map and the second feature map are similar and are shot by the same lens, and the middle frame in the selected video frame sequence is determined to be a non-switching video frame; when the similarity is greater than the preset threshold, it means that the video pictures of the first feature map and the second feature map are not similar and are shot by different lenses, and the middle frame in the selected video frame sequence is determined to be a switching video frame.

[0082] Optionally, after step S30, the recognition result of each video frame corresponding to the video to be processed is obtained; it is determined that the recognition result is not a continuous video frame of the switching video frame, as a split video, and the split video is continuously shot by the same lens. Exemplarily, the recognition result of the video frame is a switching video frame, which is represented by 1, and the recognition result of the video frame is a non-switching video frame, which is represented by 0. The recognition result of the video to be processed is 00011000111000, the 1st to 3rd frames are the first split shot, the 6th to 8th frames are the second split shot, the 12th to 14th frames are the third split shot, and the 4th to 5th frames are the switching frames, and the 9th to 11th frames are the switching frames. Optionally, the recognition result of the video to be processed can be filtered and deburred to reduce false detections. For example, 00011011000, the 6th video frame is the video frame to be processed, and after filtering and deburring, the recognition result of the video to be processed is 00011111000.

[0083] In the technical solution of this embodiment, a first feature map corresponding to each video frame in the selected video frame sequence is obtained based on a first feature extraction weight, with the first feature extraction weight decreasing sequentially based on the frame order of the selected video frame sequence; a second feature map corresponding to each video frame in the selected video frame sequence is obtained based on a second feature extraction weight, with the second feature extraction weight increasing sequentially based on the frame order of the selected video frame sequence; based on the first feature map and the second feature map, it is determined whether the intermediate frame in the selected video frame sequence is a switching video frame. Because the first feature map focuses on extracting features of video frames with a higher ranking, and the second feature map focuses on extracting features of video frames with a lower ranking, the recognition result of the intermediate frame is comprehensively determined based on the first feature map and the second feature map, thereby improving the accuracy and efficiency of identifying switching frames in the video and improving the efficiency of video segmentation and recognition.

[0084] Reference Figure 8 , Figure 8 This is a second embodiment of the video switching identification method of the present invention, based on the first embodiment, and step S30 includes:

[0085] Step S31, fusing the first feature map and the second feature map;

[0086] In step S32, the fusion result of the first feature map and the second feature map is input into a preset two-dimensional convolutional network to obtain a recognition result of whether the intermediate frame is a switching video frame.

[0087] Specifically, the first feature map and the second feature map are fused, such as Figure 3 As shown, the first feature map and the second feature map are optionally fused in the channel dimension. The fusion result of the first feature map and the second feature map is input into a preset two-dimensional convolutional network to obtain a recognition result of whether the intermediate frame is a switching video frame. Optionally, the two-dimensional convolutional network includes multiple convolutional layers, global pooling layers, fully connected layers, and softmax activation layers.

[0088] After step S32, the recognition result of each video frame corresponding to the video to be processed is obtained; it is determined that the recognition result is not a continuous video frame of the switching video frame, as a split video, the split video is continuously shot by the same lens. Exemplarily, the recognition result of the video frame is a switching video frame, which is represented by 1, and the recognition result of the video frame is a non-switching video frame, which is represented by 0. The recognition result of the video to be processed is 00011000111000, the 1st to 3rd frames are the first split shot, the 6th to 8th frames are the second split shot, the 12th to 14th frames are the third split shot, and the 4th to 5th frames are the switching frames, and the 9th to 11th frames are the switching frames. Optionally, the recognition result of the video to be processed can be filtered and deburred to reduce false detections. For example, 00011011000, the 6th video frame is the video frame to be processed, and after filtering and deburring, the recognition result of the video to be processed is 00011111000.

[0089] In the technical solution of this embodiment, the first and second feature maps are fused; the fusion result of the first and second feature maps is input into a preset two-dimensional convolutional network to obtain a result for identifying whether the intermediate frame is a switching video frame. Because the first feature map focuses on extracting features of video frames with a higher ranking, and the second feature map focuses on extracting features of video frames with a lower ranking, the recognition result of the intermediate frame is determined based on the first and second feature maps, thereby improving the accuracy and efficiency of identifying switching frames in the video and enhancing the efficiency of video segmentation and recognition.

[0090] Reference Figure 9 The present invention further provides a video switching recognition device, the video switching recognition device comprising:

[0091] A first determining module 100 is configured to obtain a first feature map corresponding to each video frame in a selected video frame sequence according to a first feature extraction weight, wherein the first feature extraction weight decreases in sequence based on a frame order of the selected video frame sequence;

[0092] A second determining module 200 is configured to obtain a second feature map corresponding to each video frame in the selected video frame sequence according to a second feature extraction weight, wherein the second feature extraction weight increases sequentially based on a frame order of the selected video frame sequence;

[0093] The identification module 300 is configured to determine whether an intermediate frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map.

[0094] In one embodiment, before obtaining the first feature map corresponding to each video frame in the selected video frame sequence according to the first feature extraction weight, the first determining module 100 is specifically configured to:

[0095] Get multiple video frames of the video to be processed;

[0096] Determine a starting frame, a window length, and a sliding step length in the multi-frame video frame, wherein the window length is an odd value;

[0097] The selected video frame sequence is acquired according to the starting frame, the window length, and the sliding step size.

[0098] In one embodiment, in determining whether the middle frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map, the identification module 300 is specifically configured to:

[0099] fusing the first feature map and the second feature map;

[0100] The fusion result of the first feature map and the second feature map is input into a preset two-dimensional convolutional network to obtain a recognition result of whether the intermediate frame is a switching video frame.

[0101] In one embodiment, in determining whether the middle frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map, the identification module 300 is specifically configured to:

[0102] comparing the similarity between the first feature map and the second feature map;

[0103] When the similarity is less than or equal to a preset threshold, determining that the middle frame in the selected video frame sequence is a non-switching video frame;

[0104] When the similarity is greater than a preset threshold, the middle frame in the selected video frame sequence is determined to be a switching video frame.

[0105] In one embodiment, after determining whether the middle frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map, the identification module 300 is specifically configured to:

[0106] Obtaining recognition results of each video frame corresponding to the video to be processed;

[0107] The continuous video frames that are non-switched video frames as the recognition results are determined as split videos, and the split videos are continuously shot by the same lens.

[0108] The present invention also provides a video switching recognition device, which includes a memory, a processor, and a video switching recognition program stored in the memory and executable on the processor. When the video switching recognition program is executed by the processor, the various steps of the video switching recognition method described in the above embodiment are implemented.

[0109] The present invention also provides a computer-readable storage medium storing a video switching recognition program. When the video switching recognition program is executed by a processor, the video switching recognition program implements the various steps of the video switching recognition method described in the above embodiment.

[0110] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0111] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, system, article, or device comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, system, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, system, article, or device comprising the element.

[0112] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment system can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, can be embodied in the form of a software product. The computer software product is stored in a computer-readable storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, parking management equipment, air conditioner, or network equipment, etc.) to execute the system described in each embodiment of the present invention.

[0113] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A video switching recognition method, characterized in that: The video switching identification method includes: Obtaining a first feature map corresponding to each video frame in the selected video frame sequence according to a first feature extraction weight, wherein the first feature extraction weight decreases in sequence based on a frame order of the selected video frame sequence; Obtaining a second feature map corresponding to each video frame in the selected video frame sequence according to a second feature extraction weight, wherein the second feature extraction weight increases sequentially based on a frame order of the selected video frame sequence; Determining whether the intermediate frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map includes: fusing the first feature map and the second feature map; inputting the fusion result of the first feature map and the second feature map into a preset two-dimensional convolutional network to obtain an identification result of whether the intermediate frame is a switching video frame.

2. The video switching recognition method according to claim 1, wherein: The first feature map corresponding to each of the video frames is determined by a first grouped convolutional network, and the second feature map corresponding to each of the video frames is determined by a second grouped convolutional network; the first grouped convolutional network and the second grouped convolutional network are both provided with convolution groups corresponding one to one to the video frames.

3. The video switching recognition method according to claim 2, wherein: In the first grouped convolutional network, the convolution group corresponding to each video frame is associated with the first feature extraction weight, and the first feature extraction weight decreases successively with the frame order of the selected video frame sequence. In the second grouped convolutional network, the convolution group corresponding to each video frame is associated with the second feature extraction weight, and the second feature extraction weight increases successively with the frame order of the selected video frame sequence.

4. The video switching recognition method according to claim 1, wherein: Before the step of obtaining the first feature map corresponding to each video frame in the selected video frame sequence according to the first feature extraction weight, the method further includes: Get multiple video frames of the video to be processed; Determine a starting frame, a window length, and a sliding step length in the multi-frame video frame, wherein the window length is an odd value; The selected video frame sequence is acquired according to the starting frame, the window length, and the sliding step size.

5. The video switching recognition method according to claim 1, wherein: The step of determining whether the middle frame in the selected video frame sequence is a switching video frame according to the first feature map and the second feature map includes: comparing the similarity between the first feature map and the second feature map; When the similarity is less than or equal to a preset threshold, determining that the middle frame in the selected video frame sequence is a non-switching video frame; When the similarity is greater than a preset threshold, the middle frame in the selected video frame sequence is determined to be a switching video frame.

6. The video switching recognition method according to claim 1, wherein: After the step of determining whether the middle frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map, the method further includes: Obtaining recognition results of each video frame corresponding to the video to be processed; The continuous video frames that are non-switched video frames as the recognition results are determined as split videos, and the split videos are continuously shot by the same lens.

7. A video switching recognition device, characterized in that: The video switching recognition device includes: A first determining module is configured to obtain a first feature map corresponding to each video frame in a selected video frame sequence according to a first feature extraction weight, wherein the first feature extraction weight decreases in sequence based on a frame order of the selected video frame sequence; a second determining module, configured to obtain a second feature map corresponding to each video frame in the selected video frame sequence according to a second feature extraction weight, wherein the second feature extraction weight increases sequentially based on a frame order of the selected video frame sequence; An identification module is used to determine whether the intermediate frame in the selected video frame sequence is a switching video frame based on the first feature map and the second feature map, including: fusing the first feature map and the second feature map; inputting the fusion result of the first feature map and the second feature map into a preset two-dimensional convolutional network to obtain an identification result of whether the intermediate frame is a switching video frame.

8. A video switching recognition device, characterized in that: The video switching identification device includes a memory, a processor, and a video switching identification program stored in the memory and executable on the processor. When the video switching identification program is executed by the processor, the various steps of the video switching identification method as described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a video switching recognition program, and when the video switching recognition program is executed by a processor, each step of the video switching recognition method according to any one of claims 1 to 6 is implemented.

Images