video behavior identification method based on image enhancement and a 3D convolutional neural network

Abstract

The invention belongs to the technical field of image processing, and particularly relates to a video behavior identification method based on image enhancement and a 3D convolutional neural network, which comprises the following steps: segmenting an input original video sequence into frames, preprocessing each frame of picture, and respectively dividing the frame of picture into a training set anda test set; Inputting the training set pictures into a behavior area enhancement network for training to obtain corresponding mask processed pictures; Training the 3D convolutional neural network byusing the picture sequence subjected to the mask processing; Inputting a test set picture to obtain a test set classification probability of the branch network; Inputting the training set pictures into a 3D convolutional neural network for training; Inputting a test set picture to obtain a test set classification probability of the branch network; And carrying out support vector machine model training on the classification probabilities of the two branch networks, and obtaining a final test set detection result. The behavior of the person in the video can be accurately identified in real time,the image information is more fully utilized, and the accuracy of behavior identification in the video is improved.

Description

technical field [0001] The invention belongs to the technical field of multimedia and computer vision, and relates to a video behavior recognition method based on image enhancement and 3D convolutional neural network. Background technique [0002] Behavior recognition is a research hotspot and cornerstone in computer vision and multimedia fields in recent years. It has broad application prospects in security, human-computer interaction, smart home and virtual reality and other fields. In actual situations, behavior recognition often uses real-time video or monitoring as the carrier to provide real-time recognition and detection of human behavior. high demands. At present, there are two main difficulties in behavior recognition: the complexity of optical flow calculation leads to poor real-time performance, and the accuracy of behavior recognition needs to be improved. [0003] At present, there are two mainstream methods of behavior recognition technology, namely the dual-...

AI Technical Summary

Problems solved by technology

Therefore, the computing requirements of the dual-stream method lead to the inability to achieve real-time recognition requirements

The 3D convolutional neural network method is often less effective than the dual-stream method under the premise of directly using the RGB information of the video for training.

If the RGB information carried by the original video is not processed, some salient features required for behavior recognition may not be collected, and the fine-grained texture information of some behavior entities cannot be well extracted and utilized.

Images