Video anomaly online detection method and device based on skeleton trajectory diffusion model

The features of video frames are extracted through the skeletal diffusion model and the predicted frames are generated, which solves the problem of insufficient detection speed and accuracy in the prior art, and achieves more efficient video anomaly detection.

CN119027445AActive Publication Date: 2024-11-26ZHEJIANG UNIV

Patent Information

Application Number
CN202411022211.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-26
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

The existing video anomaly detection technology has shortcomings in detection speed and accuracy, especially methods based on visual reconstruction and deep learning are difficult to effectively improve detection speed and accuracy.

Method used

The online detection method of video anomalies based on the skeletal diffusion model is adopted. By obtaining continuous video frame sequences, real bone posture and optical flow information are extracted, global and local features are extracted using a spatiotemporal autoencoder, and predicted spatiotemporal bones are generated based on random noise. Frame, compare real and predict frames to determine exception frames.

Benefits of technology

It significantly improves the speed and accuracy of video abnormality detection, and enables the identification of abnormal video frames more quickly and accurately.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027445A_ABST
    Figure CN119027445A_ABST
Patent Text Reader

Abstract

The invention discloses a video anomaly online detection method and device based on a skeleton trajectory diffusion model. The method comprises the following steps: acquiring a continuous video frame sequence containing a current video frame of a to-be-detected video; extracting a real skeleton posture sequence from the continuous video frame sequence by using a skeleton posture network, and tracking skeleton information in different video frames by using an optical flow neural network to obtain a real space-time skeleton sequence of the target; extracting a global feature and a local feature from the real space-time skeleton sequence, and coding the global feature and the local feature based on a space-time auto-encoder to obtain a first condition feature and a second condition feature; generating a predicted space-time skeleton frame at the current moment through inverse diffusion based on the random noise, the first condition feature and the second condition feature by adopting a skeleton trajectory diffusion model; whether the current video frame is an abnormal video frame is determined by comparing the real space-time skeleton frame and the predicted space-time skeleton frame at the current moment, so that rapid and high-precision detection can be realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of video image signal processing and machine vision, and in particular relates to a video anomaly online detection method and device based on a skeleton trajectory diffusion model. Background Art

[0002] Video anomaly detection technology is of great significance in promoting the automation and intelligence of video surveillance. The current mainstream video anomaly detection technologies can be divided into vision-based reconstruction and skeleton-based reconstruction. The basic assumption of this method is that normal video frames can be reconstructed, but abnormal video frames cannot be reconstructed. However, this type of reconstruction-based method can often reconstruct abnormal video frames well. In addition, the use of all visual information contains a lot of redundant information, which is not conducive to improving the detection accuracy and detection speed.

[0003] The patent application with publication number CN107222659A discloses a video anomaly detection method. First, according to the characteristics of local consistency of video jitter, combined with the idea of ​​block division, the grayscale projection algorithm is used to complete the global motion estimation, and then the frame sequence that can represent the video jitter is selected as the valid frame according to the longest path algorithm. The jitter parameters such as the jitter rate, jitter frequency and jitter amplitude of the video jitter are calculated according to the global motion parameters of the valid frame, and the influencing factors are set for the jitter parameters. Finally, the jitter degree of the video is calculated by the weighted average method.

[0004] With the development of deep learning, people often perform target video anomaly detection based on deep learning. For example, a patent application with publication number CN113435432A discloses a video anomaly detection model training method, including: obtaining a training video frame sequence, corresponding sequence feature information, and target feature information of a training screening video frame corresponding to the training video frame sequence; inputting the training video frame sequence into an initial video anomaly detection model to obtain predicted feature information corresponding to the training screening video frame; generating temporal positive training samples of a temporal discrimination model corresponding to the initial video anomaly detection model based on the target feature information and the sequence feature information, and generating temporal negative training samples based on the sequence feature information and the predicted feature information; performing adversarial training on the initial video anomaly detection model and the temporal discrimination model based on the target feature information, the predicted feature information, the temporal positive training samples, and the temporal negative training samples until the convergence condition is met to obtain a target video anomaly detection model.

[0005] Although the above two technical solutions can realize the detection of video anomalies, the detection speed needs to be improved. Summary of the invention

[0006] In view of the above, an object of the present invention is to provide a method and device for online detection of video anomalies based on a skeleton trajectory diffusion model to improve the accuracy and speed of video anomaly detection.

[0007] To achieve the above-mentioned object of the invention, an embodiment of the present invention provides a video anomaly online detection method based on a skeleton trajectory diffusion model, comprising the following steps:

[0008] Obtain a continuous video frame sequence including the current video frame of the video to be detected;

[0009] The skeleton pose network is used to extract the real skeleton pose sequence from the continuous video frame sequence, and the optical flow neural network is used to track the skeleton information in different video frames to obtain the real spatiotemporal skeleton sequence of the target;

[0010] Extracting global features and local features from a real spatiotemporal skeleton sequence, and encoding the global features and local features based on a spatiotemporal autoencoder to obtain a first conditional feature and a second conditional feature;

[0011] The skeleton trajectory diffusion model is used to generate a predicted spatiotemporal skeleton frame at the current moment through reverse diffusion based on random noise, the first conditional feature and the second conditional feature;

[0012] By comparing the real spatiotemporal skeleton frame at the current moment with the predicted spatiotemporal skeleton frame, it is determined whether the current video frame is an abnormal video frame.

[0013] Preferably, extracting global features and local features from real spatiotemporal skeleton sequences includes:

[0014] Extract the motion trend of human skeleton in the whole video frame from the real spatiotemporal skeleton sequence as the global feature;

[0015] The motion changes of human posture between video frames are extracted from the real spatiotemporal skeleton sequence as local features.

[0016] Preferably, the encoder and decoder in the spatiotemporal autoencoder adopt a Unet model, and the Unet model is used to learn the global features in the latent space to obtain the first conditional features, and the Unet model is used to learn the local features in the latent space to obtain the second conditional features.

[0017] Preferably, the bone trajectory diffusion model is used to generate a predicted spatiotemporal bone sequence through reverse diffusion based on random noise, the first conditional feature and the second conditional feature, including:

[0018] Random noise is used as the initial noise, and the concatenation of the first conditional feature and the second conditional feature is used as the guiding body condition for each time step. Multi-step denoising is performed to generate a predicted spatiotemporal skeleton sequence.

[0019] Preferably, the bone trajectory diffusion model is used to generate a predicted spatiotemporal bone sequence through reverse diffusion based on random noise, the first conditional feature and the second conditional feature, including:

[0020] Using random noise as initial noise and the first conditional feature as the guiding body condition of each time step, multi-step denoising is performed to generate the first predicted candidate spatiotemporal skeleton sequence;

[0021] Using random noise as initial noise and the second conditional feature as the guiding body condition of each time step, multi-step denoising is performed to generate a second prediction candidate spatiotemporal skeleton sequence;

[0022] The first predicted candidate spatiotemporal skeleton sequence and the second predicted candidate spatiotemporal skeleton sequence are merged and fused to obtain a predicted spatiotemporal skeleton sequence.

[0023] Preferably, determining whether the current video frame is an abnormal video frame by comparing the real spatiotemporal skeleton frame at the current moment with the predicted spatiotemporal skeleton frame includes:

[0024] Calculate the anomaly scores of the real spatiotemporal skeleton frame at the current moment and the predicted spatiotemporal skeleton frame at the current moment. When the anomaly score is higher than a preset threshold, the current video frame is identified as an abnormal video frame.

[0025] Preferably, the anomaly scores of the real spatiotemporal skeleton frame and the predicted spatiotemporal skeleton frame are calculated by L1 norm.

[0026] To achieve the above-mentioned purpose of the invention, the embodiment further provides a video anomaly online detection device based on a skeleton trajectory diffusion model, comprising:

[0027] A data acquisition module, which is used to acquire a continuous video frame sequence including a current video frame of the video to be detected;

[0028] A skeleton extraction module is used to extract a real skeleton posture sequence from a continuous video frame sequence using a skeleton posture network, and to track the skeleton information in different video frames using an optical flow neural network to obtain a real spatiotemporal skeleton sequence of the target;

[0029] A feature encoding module, which is used to extract global features and local features from the real spatiotemporal skeleton sequence, and encode the global features and local features based on the spatiotemporal autoencoder to obtain a first conditional feature and a second conditional feature;

[0030] A spatiotemporal skeleton frame prediction module, which is used to generate a predicted spatiotemporal skeleton frame at the current moment through reverse diffusion based on random noise, the first conditional feature and the second conditional feature by using a skeleton trajectory diffusion model;

[0031] The abnormal confirmation module is used to determine whether the current video frame is an abnormal video frame by comparing the real space-time skeleton frame and the predicted space-time skeleton frame at the current moment.

[0032] To achieve the above-mentioned purpose of the invention, an embodiment also provides a computing device, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the above-mentioned video anomaly online detection method based on the skeleton trajectory diffusion model.

[0033] To achieve the above-mentioned purpose of the invention, an embodiment further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the above-mentioned video anomaly online detection method based on the skeleton trajectory diffusion model is implemented.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] Based on the spatiotemporal skeleton sequence extracted from the continuous video frame sequence, a guiding condition containing global features and local features is constructed, and a diffusion model is used to generate a predicted spatiotemporal skeleton frame based on the guiding condition. Then, based on the real spatiotemporal skeleton frame and the predicted spatiotemporal skeleton frame at the current moment, it is determined whether the current video frame is an abnormal video frame. This can greatly improve the detection speed and also improve the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 is a flow chart of a method for online video anomaly detection based on a skeleton trajectory diffusion model provided by an embodiment;

[0038] Figure 2 is a training schematic diagram of a skeleton trajectory diffusion model provided in an embodiment;

[0039] Figure 3 It is a structural schematic diagram of a video anomaly online detection device based on a skeleton trajectory diffusion model provided in an embodiment. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0041] like Figure 1As shown, the embodiment provides a video anomaly online detection method based on a skeleton trajectory diffusion model, comprising the following steps:

[0042] S1, obtaining a continuous video frame sequence including a current video frame of the video to be detected.

[0043] In an embodiment, the obtained continuous video frame sequence includes a current video frame and also includes historical video frames other than the current video frame. These video frame sequences are subsequently used to extract skeleton information to construct a guidance vector.

[0044] S2, uses the skeleton pose network to extract the real skeleton pose sequence from the continuous video frame sequence, and uses the optical flow neural network to track the skeleton information in different video frames to obtain the real spatiotemporal skeleton sequence of the target.

[0045] In an embodiment, the skeleton posture network may adopt the Alpha Pose model, and the optical flow neural network may adopt FlowNet.

[0046] S3, extracting global features and local features from the real spatiotemporal skeleton sequence, and encoding the global features and local features based on the spatiotemporal autoencoder to obtain the first conditional features and the second conditional features.

[0047] In an embodiment, the movement trend of the human skeleton in the entire video frame is extracted from the real space-time skeleton sequence as a global feature; the movement change of the human posture between video frames is extracted from the real space-time skeleton sequence as a local feature, and then the global feature and the local feature are encoded using a space-time autoencoder to obtain a first conditional feature and a second conditional feature.

[0048] In the embodiment, the encoder and decoder in the spatiotemporal autoencoder adopt the Unet model, and the Unet model is used to learn the global features in the latent space to obtain the first conditional features, and the Unet model is used to learn the local features in the latent space to obtain the second conditional features.

[0049] S4, using the skeleton trajectory diffusion model to generate the predicted spatiotemporal skeleton frame at the current moment through reverse diffusion based on random noise, the first conditional feature and the second conditional feature.

[0050] In the embodiment, the structure of the bone trajectory diffusion model is as follows Figure 2 As shown, before being applied, the skeleton trajectory diffusion model is trained using a public video anomaly detection dataset.

[0051] In the bone trajectory diffusion model, according to the Markov chain derivation rule, x t represents the image of the t-th diffusion step, so the forward diffusion equation can be expressed as:

[0052]

[0053] Among them, q(x t |x t-1 ) is a normal distribution, and is the mean of the Gaussian function, β t is the variance of the Gaussian function.

[0054] In the inverse diffusion process, the model needs to learn and represent the conditional probability distribution p θ (x t-1 |x t )

[0055] p θ (x t-1 |x t )=N(x t-1 ;μ θ (x t ,t),∑ θ (x t ,t))

[0056] The mean and variance also depend on the noise level. The model needs to represent these means and variances through learning. We choose to keep the variance fixed and only learn the mean of this conditional probability distribution. t is the spatiotemporal skeleton sequence sampled from the dataset, and t represents the tth step in the diffusion process.

[0057] In one method, a bone trajectory diffusion model is used to generate a predicted spatiotemporal bone sequence through reverse diffusion based on random noise, a first conditional feature and a second conditional feature, including: using random noise as the initial noise, and using the concatenation of the first conditional feature and the second conditional feature as the guide body condition for each time step, performing multi-step denoising, and generating a predicted spatiotemporal bone sequence.

[0058] In another method, a bone trajectory diffusion model is used to generate a predicted spatiotemporal bone sequence through reverse diffusion based on random noise, first conditional features and second conditional features, including: using random noise as the initial noise and the first conditional features as the guiding body conditions for each time step, performing multi-step denoising to generate a first predicted candidate spatiotemporal bone sequence; using random noise as the initial noise and the second conditional features as the guiding body conditions for each time step, performing multi-step denoising to generate a second predicted candidate spatiotemporal bone sequence; merging and fusing the first predicted candidate spatiotemporal bone sequence and the second predicted candidate spatiotemporal bone sequence to obtain a predicted spatiotemporal bone sequence.

[0059] S5, determining whether the current video frame is an abnormal video frame by comparing the real spatiotemporal skeleton frame at the current moment with the predicted spatiotemporal skeleton frame.

[0060] In an embodiment, when making an abnormality judgment, the abnormality scores of the real space-time skeleton frame at the current moment and the predicted space-time skeleton frame at the current moment are calculated, where the abnormality score can be calculated using the L1 norm, that is, the L1 norm of the real space-time skeleton frame and the predicted space-time skeleton frame is calculated as the abnormality score. When the abnormality score is higher than a preset threshold, the current video frame is determined to be an abnormal video frame.

[0061] like Figure 3 As shown, the embodiment also provides a video anomaly online detection device based on the bone trajectory diffusion model, including a data acquisition module 31, a bone extraction module 32, a feature encoding module 33, a predicted space-time bone frame module 34, and an abnormality confirmation module 35, wherein the data acquisition module 31 is used to obtain a continuous video frame sequence including a current video frame of the video to be detected; the bone extraction module 32 is used to extract a real bone posture sequence from the continuous video frame sequence using a bone posture network, and use an optical flow neural network to track the bone information in different video frames to obtain a real space-time bone sequence of the target; the feature encoding module 33 is used to extract global features and local features from the real space-time bone sequence, and encode the global features and local features based on the space-time autoencoder to obtain a first conditional feature and a second conditional feature; the predicted space-time bone frame module 34 is used to generate a predicted space-time bone frame at the current moment through inverse diffusion based on random noise, the first conditional feature and the second conditional feature using the bone trajectory diffusion model; the abnormality confirmation module 35 is used to determine whether the current video frame is an abnormal video frame by comparing the real space-time bone frame at the current moment with the predicted space-time bone frame.

[0062] It should be noted that the video anomaly online detection device based on the skeleton trajectory diffusion model provided in the above embodiment should be illustrated by the division of the above-mentioned functional modules when performing video anomaly online detection. The above-mentioned function allocation can be completed by different functional modules as needed, that is, the internal structure of the terminal or server is divided into different functional modules to complete all or part of the functions described above. In addition, the video anomaly online detection device based on the skeleton trajectory diffusion model provided in the above embodiment and the video anomaly online detection construction method embodiment based on the skeleton trajectory diffusion model belong to the same concept. The specific implementation process is detailed in the video anomaly online detection method embodiment based on the skeleton trajectory diffusion model, which will not be repeated here.

[0063] Based on the same inventive concept, an embodiment further provides a computing device, including a memory and one or more processors, wherein executable code is stored in the memory, and when the one or more processors execute the executable code, the method for online video anomaly detection based on the skeleton trajectory diffusion model is implemented, specifically including the following steps:

[0064] S1, obtaining a continuous video frame sequence including a current video frame of the video to be detected;

[0065] S2, using the skeleton posture network to extract the real skeleton posture sequence from the continuous video frame sequence, using the optical flow neural network to track the skeleton information in different video frames, and obtain the real spatiotemporal skeleton sequence of the target;

[0066] S3, extracting global features and local features from the real spatiotemporal skeleton sequence, and encoding the global features and local features based on the spatiotemporal autoencoder to obtain the first conditional features and the second conditional features;

[0067] S4, using the skeleton trajectory diffusion model to generate a predicted spatiotemporal skeleton frame at the current moment through reverse diffusion based on random noise, the first conditional feature and the second conditional feature;

[0068] S5, determining whether the current video frame is an abnormal video frame by comparing the real spatiotemporal skeleton frame at the current moment with the predicted spatiotemporal skeleton frame.

[0069] The computing device provided in the embodiment, in addition to the processor and memory, also includes hardware required for other services such as internal bus, network interface, memory, etc. at the hardware level. The memory is a non-volatile memory, and the processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the online video anomaly detection method based on the skeleton trajectory diffusion model described in S1-S5 above. Of course, in addition to the software implementation, the present invention does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0070] Based on the same inventive concept, an embodiment further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the above-mentioned video anomaly online detection method based on the skeleton trajectory diffusion model is implemented, which specifically includes the following steps:

[0071] S1, obtaining a continuous video frame sequence including a current video frame of the video to be detected;

[0072] S2, using the skeleton posture network to extract the real skeleton posture sequence from the continuous video frame sequence, using the optical flow neural network to track the skeleton information in different video frames, and obtain the real spatiotemporal skeleton sequence of the target;

[0073] S3, extracting global features and local features from the real spatiotemporal skeleton sequence, and encoding the global features and local features based on the spatiotemporal autoencoder to obtain the first conditional features and the second conditional features;

[0074] S4, using the skeleton trajectory diffusion model to generate a predicted spatiotemporal skeleton frame at the current moment through reverse diffusion based on random noise, the first conditional feature and the second conditional feature;

[0075] S5, determining whether the current video frame is an abnormal video frame by comparing the real spatiotemporal skeleton frame at the current moment with the predicted spatiotemporal skeleton frame.

[0076] In the embodiment, computer-readable media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data.

[0077] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A video anomaly online detection method based on a skeleton trajectory diffusion model, characterized in that: The following steps are involved: Obtain a continuous video frame sequence including the current video frame of the video to be detected; The skeleton pose network is used to extract the real skeleton pose sequence from the continuous video frame sequence, and the optical flow neural network is used to track the skeleton information in different video frames to obtain the real spatiotemporal skeleton sequence of the target; Extracting global features and local features from a real spatiotemporal skeleton sequence, and encoding the global features and local features based on a spatiotemporal autoencoder to obtain a first conditional feature and a second conditional feature; The bone trajectory diffusion model is used to generate a predicted spatiotemporal bone frame at the current moment through reverse diffusion based on random noise, the first conditional feature and the second conditional feature; By comparing the real spatiotemporal skeleton frame at the current moment with the predicted spatiotemporal skeleton frame, it is determined whether the current video frame is an abnormal video frame.

2. The video anomaly online detection method based on the skeleton trajectory diffusion model according to claim 1 is characterized in that: Extract global and local features from real spatiotemporal skeleton sequences, including: Extract the motion trend of human skeleton in the whole video frame from the real spatiotemporal skeleton sequence as the global feature; The motion changes of human posture between video frames are extracted from the real spatiotemporal skeleton sequence as local features.

3. The video anomaly online detection method based on the skeleton trajectory diffusion model according to claim 1 is characterized in that: The encoder and decoder in the spatiotemporal autoencoder adopt the Unet model. The Unet model is used to learn the global features in the latent space to obtain the first conditional features, and the Unet model is used to learn the local features in the latent space to obtain the second conditional features.

4. The video anomaly online detection method based on the skeleton trajectory diffusion model according to claim 1 is characterized in that: The bone trajectory diffusion model is used to generate a predicted spatiotemporal bone sequence based on random noise, the first conditional feature and the second conditional feature through reverse diffusion, including: Using random noise as the initial noise and the concatenation of the first conditional feature and the second conditional feature as the guiding body condition for each time step, multi-step denoising is performed to generate a predicted spatiotemporal skeleton sequence.

5. The video anomaly online detection method based on the skeleton trajectory diffusion model according to claim 1 is characterized in that: The bone trajectory diffusion model is used to generate a predicted spatiotemporal bone sequence based on random noise, the first conditional feature and the second conditional feature through reverse diffusion, including: Using random noise as initial noise and the first conditional feature as the guiding body condition of each time step, multi-step denoising is performed to generate the first predicted candidate spatiotemporal skeleton sequence; Using random noise as initial noise and the second conditional feature as the guiding body condition of each time step, multi-step denoising is performed to generate a second prediction candidate spatiotemporal skeleton sequence; The first predicted candidate spatiotemporal skeleton sequence and the second predicted candidate spatiotemporal skeleton sequence are merged and fused to obtain a predicted spatiotemporal skeleton sequence.

6. The video anomaly online detection method based on the skeleton trajectory diffusion model according to claim 1 is characterized in that: By comparing the real spatiotemporal skeleton frame at the current moment with the predicted spatiotemporal skeleton frame, the current video frame is determined to be an abnormal video frame, including: The anomaly score between the real spatiotemporal skeleton frame at the current moment and the predicted spatiotemporal skeleton frame at the current moment is calculated. When the anomaly score is higher than a preset threshold, the current video frame is identified as an abnormal video frame.

7. The video anomaly online detection method based on the skeleton trajectory diffusion model according to claim 6 is characterized in that: The anomaly scores of the real spatiotemporal skeleton frames and the predicted spatiotemporal skeleton frames are calculated using the L1 norm.

8. A video anomaly online detection device based on a skeleton trajectory diffusion model, characterized in that: include: A data acquisition module, which is used to acquire a continuous video frame sequence including a current video frame of the video to be detected; A skeleton extraction module is used to extract a real skeleton posture sequence from a continuous video frame sequence using a skeleton posture network, and to track the skeleton information in different video frames using an optical flow neural network to obtain a real spatiotemporal skeleton sequence of the target; A feature encoding module, which is used to extract global features and local features from the real spatiotemporal skeleton sequence, and encode the global features and local features based on the spatiotemporal autoencoder to obtain a first conditional feature and a second conditional feature; A predicted spatiotemporal skeleton frame module, which is used to generate a predicted spatiotemporal skeleton frame at the current moment through reverse diffusion based on random noise, the first conditional feature and the second conditional feature by using a skeleton trajectory diffusion model; The abnormal confirmation module is used to determine whether the current video frame is an abnormal video frame by comparing the real space-time skeleton frame and the predicted space-time skeleton frame at the current moment.

9. A computing device comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the one or more processors execute the executable code, they are used to implement the video anomaly online detection method based on the skeleton trajectory diffusion model according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, the video anomaly online detection method based on a skeleton trajectory diffusion model described in any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Conditional diffusion model-based compressed domain video anomaly detection method

    CN117376575A

  • Traffic video anomaly detection method and system based on dual-condition diffusion model

    CN117789087A

  • Human body video generation method based on time sequence consistent hidden space guide diffusion model

    CN117994708A

Cited By

  • Catenary component thermal defect detection method based on space-time fusion and vehicle speed normalization

    CN121456768A