Video motion identification method based on time domain segmentation network

Abstract

The present invention provides a video motion identification method based on a time domain segmentation network. The main content of the method comprises: time domain segmentation network (TSN) based on segmented sampling, clustering function and analysis, time domain segmentation network input and training strategy and motion identification in non-cutting video. The process of the method comprises: dividing video into equal continuous time, randomly extracting one segment from corresponding segments, each segment in a sequence generating motion-level segment-stage prediction, designing a common function, aggregating the segment-level prediction into a video-level score, and in the training process, defining an optimal target in the video-level prediction, and performing optimization through an iteration updating model parameters. The sampling and aggregation modules based on segmentation are employed to establish a long-term time structure to use the whole motion video to effectively learn a motion model, store long-term video and improve the sensitivity and accuracy for the detection and identification of the motion.

Description

technical field [0001] The invention relates to the field of action recognition, in particular to a video action recognition method based on a time-domain segmentation network. Background technique [0002] With the rapid development of science and technology and the progress of society, people use video acquisition technology in all aspects of their daily life. However, after people have acquired a large amount of video data, they often need to artificially watch, identify and mark actions in the video. Therefore, video action recognition technology has attracted more and more people's attention, and its application scope has become wider and wider, such as identifying and analyzing the behavior of suspicious people in surveillance videos of vending machines, ATM machines, shopping malls, stations and other public places, basketball games, etc. Shooting action analysis, dance video analysis for practice, identification and detection of dangerous actions of drivers in road d...

AI Technical Summary

Problems solved by technology

However, training deep convolutional neural networks in traditional methods requires large training samples, but the data resources in this area are limited, and its limited storage space severely limits the duration of the video, which will cause the video to lose important information

Images