CN-121985096-A - Monitoring video compression storage method

CN121985096ACN 121985096 ACN121985096 ACN 121985096ACN-121985096-A

Abstract

The application relates to the technical field of image communication transmission, in particular to a monitoring video compression storage method, which comprises the steps of obtaining a frame sequence of a monitoring video and constructing a three-dimensional video tensor, calculating time sequence disorder degree of any coordinate position and structural consistency of the coordinate position for any coordinate position in any frame image, calculating self-adaptive weights of all coordinate positions based on local structural consistency, constructing a self-adaptive weight matrix, wherein all weight values are inversely proportional to the local structural consistency of corresponding coordinate positions, processing the three-dimensional video tensor by adopting a weighted low-rank tensor decomposition algorithm to obtain a low-rank background tensor and a sparse front Jing Zhangliang, wherein the self-adaptive weight matrix is used for applying space variable sparsity constraint to the sparse foreground tensor, and respectively compressing and storing the low-rank background tensor and the sparse front Jing Zhangliang. The application has the effect of improving the image compression efficiency and the fidelity.

Inventors

BAI RIJIAN
NI JIANZHONG
XU XIJIAN
Zhou Yaocai

Assignees

广州唯邦特种车辆有限公司

Dates

Publication Date: 20260505
Application Date: 20260203

Claims (10)

1. The monitoring video compression storage method is characterized by comprising the steps of acquiring a frame sequence of a monitoring video and constructing a three-dimensional video tensor; for any coordinate position in any frame of image, extracting a gray value sequence of a corresponding pixel point in the three-dimensional video tensor in a preset analysis time window, and calculating time sequence disturbance degree based on the change of gray values in the gray value sequence; For any coordinate position, constructing a local neighborhood window, calculating the energy concentration according to the time sequence disorder degree difference of each coordinate position in the local neighborhood window, and taking the product of the energy concentration and the time sequence disorder degree corresponding to the coordinate position as the structural consistency of the coordinate position; calculating self-adaptive weights of the coordinate positions based on the local structure consistency, and constructing a self-adaptive weight matrix, wherein each weight value is inversely proportional to the local structure consistency of the corresponding coordinate position; and processing the three-dimensional video tensor by adopting a weighted low-rank tensor decomposition algorithm to obtain a low-rank background tensor and a sparse front Jing Zhangliang, wherein the adaptive weight matrix is used for applying space variable sparsity constraint on the sparse foreground tensor, and respectively compressing and storing the low-rank background tensor and the sparse front Jing Zhangliang.
2. The method according to claim 1, wherein the step of calculating the timing disturbance based on the change of the gradation values in the gradation value sequence includes taking, as an adjacent difference, an absolute difference between the gradation value and a gradation value corresponding to a previous frame image for any one of the gradation values in the gradation value sequence, taking, as a deviation difference, a difference between the gradation value and a gradation value average value in the gradation value sequence, taking, as a local difference, a product of a result of smoothing the deviation difference and the adjacent difference, and taking, as the timing disturbance, an average value of a plurality of local differences corresponding to the gradation sequence.
3. The method of claim 1, wherein the step of calculating the energy concentration from the difference in the temporal turbulence level for each coordinate location in the local neighborhood window comprises using as the energy concentration a ratio of the square of the sum of the temporal turbulence level values for each coordinate location in the local neighborhood window to the sum of the squares of all the temporal turbulence level values.
4. A surveillance video compression storage method according to claim 3, wherein for any coordinate position, the local neighborhood window is a square area of 3 x 3 pixels centered on that coordinate position.
5. The method for storing the surveillance video compression according to claim 1, wherein the step of calculating the adaptive weight of each coordinate position based on the local structural coherence includes taking, as the adjustment degree, a sum of the local structural coherence, the square root of the local structural coherence, and a preset constant of the coordinate position, and taking, as the adaptive weight of the coordinate position, a ratio of a preset weight upper limit value to the adjustment degree.
6. The method of claim 1, further comprising performing a Gaussian filter process on a sequence of frames of the surveillance video prior to constructing the three-dimensional video tensor.
7. The surveillance video compression storage method of claim 1, wherein the analysis time window comprises 50 to 100 consecutive video frames.
8. The method for compressed storage of surveillance video according to claim 5, wherein the preset weight upper limit value is a maximum local structure consistency value of the coordinate position within an analysis time window.
9. The method of claim 1, wherein the step of applying a spatially variable sparsity constraint to the sparse foreground tensor comprises performing a Hadamard operation on an adaptive weight matrix and a sparse front Jing Zhangliang.
10. The method of claim 6, wherein the step of performing a Gaussian filter process on the frame sequence of the surveillance video comprises converting video frames of the RGB color space into the YCbCr color space and extracting a luminance component as a gray scale value, and performing the Gaussian filter process on the gray scale value.

Description

Monitoring video compression storage method Technical Field The application relates to the technical field of image communication transmission, in particular to a monitoring video compression storage method. Background With the development of smart cities and security systems, video monitoring has become a core means for guaranteeing public security. The monitoring camera usually needs to record continuously all weather, and a large amount of background redundant information which is static for a long time is contained in a large amount of video data generated by the monitoring camera, so that huge cost pressure is brought to storage equipment and network transmission, and the retrieval efficiency of key events is reduced. To solve the above problem, a Weighted Low-rank tensor decomposition (Weighted Low-Rank Tensor Decomposition, WLRTD) algorithm is commonly used in the prior art. The algorithm constructs the video data into the three-dimensional tensor, and separates the three-dimensional tensor into the low-rank component representing the background and the sparse component representing the foreground target by utilizing the tensor decomposition theory, so that the high-dimensional structural information of the video data can be better kept compared with the traditional method. However, in the actual monitoring scene, there are complicated dynamic background interferences such as abrupt illumination, tree shadow shake, rain and snow weather, etc. The original WLRTD algorithm has inherent drawbacks in processing such video, which typically employs a fixed global weight parameter to constrain the sparsity of the foreground, which makes it difficult to adaptively distinguish real moving objects from dynamic background noise. The method can easily misjudge the dynamic background as the foreground target by the algorithm or lose the details of the foreground target due to excessive inhibition, and finally, the separation precision of the background and the foreground is not high, so that the reconstruction quality and the subsequent evidence obtaining value of the compressed video are affected. Disclosure of Invention In order to improve the separation precision between the foreground and the background in the video compression process, the application provides a monitoring video compression storage method. The application provides a monitoring video compression storage method, which adopts the following technical scheme: the monitoring video compression storage method comprises the steps of obtaining a frame sequence of a monitoring video and constructing a three-dimensional video tensor; for any coordinate position in any frame of image, extracting a gray value sequence of a corresponding pixel point in the three-dimensional video tensor in a preset analysis time window, and calculating time sequence disturbance degree based on the change of gray values in the gray value sequence; For any coordinate position, constructing a local neighborhood window, calculating the energy concentration according to the time sequence disorder degree difference of each coordinate position in the local neighborhood window, and taking the product of the energy concentration and the time sequence disorder degree corresponding to the coordinate position as the structural consistency of the coordinate position; calculating self-adaptive weights of the coordinate positions based on the local structure consistency, and constructing a self-adaptive weight matrix, wherein each weight value is inversely proportional to the local structure consistency of the corresponding coordinate position; and processing the three-dimensional video tensor by adopting a weighted low-rank tensor decomposition algorithm to obtain a low-rank background tensor and a sparse front Jing Zhangliang, wherein the adaptive weight matrix is used for applying space variable sparsity constraint on the sparse foreground tensor, and respectively compressing and storing the low-rank background tensor and the sparse front Jing Zhangliang. The gray level fluctuation of the real foreground region can cause large gray level change in the video frame due to motion, and the change generated by light disturbance and slight camera shake is more similar to low-frequency smooth disturbance, and based on the change, the gray level value sequence of the pixels in the analysis time window is extracted and the time sequence disturbance degree is calculated, so that the follow-up processing can distinguish the moving object from the non-real foreground region. And finally, constructing a self-adaptive weight matrix based on the local structural consistency, so as to improve the separation precision of the foreground and the background, avoid the excessive sparsification of the foreground or the false separation of the background, and improve the separation precision of the foreground and the background compared with the sparsity weight of the constraint foreground fixed in the prio