Motion recognition method

Abstract

The invention provides an action recognition method, and the method comprises the following steps: segmenting an action video sample, and calculating a dynamic image of each video clip; inputting the dynamic and static images of each video clip into a feature extractor; extracting a motion feature vector of the dynamic image; constructing a feature center group to obtain histogram expression; inputting the histogram expression into a histogram connection layer to obtain a complete histogram expression of the action video sample; inputting a multi-layer perceptron to form a motion feature quantization network; performing training to converge; inputting the dynamic and static images into a feature extractor, an enhancer and a soft quantizer in the trained motion feature quantization network to obtain histogram expression; inputting the histogram expression into a significant motion feature extractor to obtain a significant motion feature map; inputting the significant motion feature map into a convolutional neural network to form an action classifier; performing training to converge; calculating a dynamic image and a static image of each fragment of the test action video sample, and inputting the dynamic image and the static image into the trained action classifier to realize action recognition.

Description

technical field [0001] The invention relates to an action recognition method, which belongs to the technical field of action recognition. Background technique [0002] Action recognition is an important topic in the field of computer vision, and it has a wide range of applications in video surveillance, behavior analysis, human-computer interaction and other fields. Although bone-based action recognition has received more and more attention due to the development of cheap depth cameras, these methods are limited by the accuracy of depth cameras. will go wrong. Compared with depth cameras, RGB devices are more mature and reliable. Therefore, many scholars study action recognition based on RGB videos. [0003] Most existing methods perform action recognition by extracting image-level features of video frames. These methods are not dedicated to extracting motion features of actions in videos. However, for video analysis, it is very important to obtain dynamic information. Mot...

AI Technical Summary

Problems solved by technology

Although bone-based action recognition has received more and more attention due to the development of cheap depth cameras, these methods are limited by the accuracy of depth cameras. will go wrong

Images