Abnormal behavior detection method based on improved pseudo three-dimensional residual neural network

Abstract

The invention discloses an abnormal behavior detection method based on an improved pseudo three-dimensional residual neural network, and the method comprises the following steps of firstly, dividing each video in a training set into a plurality of video clips; secondly, inputting all video clips of the video into the improved pseudo three-dimensional residual neural network to obtain the characteristics of the video clips; secondly, taking the average value of the feature vectors of all frames in each segment, and then performing L2 norm normalization on the feature average value to obtain the feature vector of the video segment; and finally, inputting the feature vector of the video clip into a three-layer full connection neural network, and outputting an abnormal score of the video clip. Experimental results show that compared with an existing method, the method provided by the invention further improves the accuracy of the abnormal behavior detection and is more suitable for the practical application.

Description

technical field [0001] The invention relates to an abnormal behavior detection method in a monitoring video scene, in particular to an abnormal behavior detection method based on multi-instance learning and improved pseudo three-dimensional residual neural network, and belongs to the technical field of video analysis. Background technique [0002] Traditional video surveillance mainly relies on artificial monitoring of abnormal behaviors in the scene, which not only requires extremely high labor costs, but also easily causes visual fatigue, and even causes some abnormal behaviors to not be observed in time; abnormal behavior detection and analysis aims to The abnormal behavior in the monitoring scene is automatically detected through video signal processing and machine learning algorithms, so as to help people take corresponding measures in a timely manner; therefore, the detection of abnormal behavior in the monitoring scene has very important research significance. [0003...

AI Technical Summary

Problems solved by technology

[0002] Traditional video surveillance mainly relies on artificial monitoring of abnormal behaviors in the scene, which not only requires extremely high labor costs, but also easily causes visual fatigue, and even causes some abnormal behaviors to not be observed in time; abnormal behavior detection and analysis aims to The abnormal behavior in the monitoring scene is automatically detected through video signal processing and machine learning algorithms, so as to help people take corresponding measures in a timely manner; therefore, the abnormal behavior detection of the monitoring scene has very important research significance

[0003] Early research work on abnormal behavior detection used low-level trajectory features to describe normal patterns. However, due to the difficulty of obtaining reliable trajectories, these methods are not robust in complex or crowded scenes with multiple occlusions; considering trajectory features And the lack of low-level spatio-temporal features, histogram of oriented gradient (HOG), histogram of optical flow (HOF) and boundary histogram (MBH) are widely used. On this basis, Markov random field model (MRF), social Force Model (SFM), Multiscale Optical Flow Histogram (MHOF), and Hybrid Dynamic Texture (MDT) have been proposed successively; these methods model normal behavior based on training data of normal behavior and detect low-probability patterns as anomalies, however, These artificially designed features are difficult to effectively reflect behavioral characteristics, and the calculation is complex

[0004] With the success of sparse representation and dictionary learning methods on some computer vision problems, researchers began to use sparse representation to learn dictionaries of normal behavior. During testing, patterns with large reconstruction errors were considered abnormal behaviors; Recently, some researchers have used deep learning-based autoencoders to learn normal behavior models and use reconstruction loss to detect anomalies; the methods are based on the assumption that any behavior that deviates from the learned normal behavior patterns will be regarded as abnormal; However, this assumption may not hold because both normal and abnormal behaviors are complex and diverse, and the line between them is sometimes blurred

Images