Multiple dimensioned convolution neural network-based real time human body abnormal behavior identification method

Abstract

The invention discloses a multiple dimensioned convolution neural network-based real time human body abnormal behavior identification method. A convolution neural network is used for replacing a conventional feature extraction algorithm, and the convolution neural network is improved so as to satisfy requirements for human body behavior classification; specifically, three dimensional convolution, three dimensional down-sampling, NIN, three dimensional pyramid structures are added; human body abnormal behavior feature extraction capability of the convolution neural network is enabled to be increased; training operation is performed in a specific video set, features with great classification capacity can be obtained, robustness and accuracy of a whole identification algorithm can be improved, GPU speed is increased so as to satisfy requirements for practical application, and therefore multi-channel videos can be monitored in real time.

Description

technical field [0001] The invention relates to the fields of computer vision and machine learning, in particular to the detection technology of abnormal human behavior in videos. Background technique [0002] Railway stations, banks, airports and other occasions with high safety requirements have a huge demand for human behavior recognition. If the system can recognize human behavior, it can automatically determine abnormal situations and call the police, thereby greatly reducing labor costs and improving monitoring. , to achieve online monitoring, real-time alarm. [0003] Traditional human behavior recognition schemes are basically based on background modeling and feature matching. This type of scheme consists of three steps: the first step is to mainly extract spatiotemporal feature points, that is, pixels with temporal and spatial characteristics, and use background difference Or the optical flow method for background modeling and foreground extraction; the second step...

AI Technical Summary

Problems solved by technology

[0006] 1. In the Gaussian background modeling method, if the color of the foreground and the background are the same or similar, when judging the foreground and finding connected areas, it is easy to misjudge part of the foreground as the background, resulting in the loss of the extracted area;

[0007] 2. Use the method of extracting the features of the pictures and then comparing them, which cannot make good use of the characteristics of the time domain to a certain extent;

[0008] 3. The template matching method is used in behavior recognition. Due to the difference between the feature distribution of the training set video and the real monitoring environment, it cannot well identify some situations that are different from the samples in the sample library, such as changes in video angles and behaviors. variation, prone to false positives and false negatives

Images