Training method of human action recognition and recognition method

Abstract

The invention provides a training method of human action recognition, comprising the following steps: extracting space-time interest points from a video file; quantizing all the space-time interest points to corresponding video words according to the feature descriptors contained by the space-time interest points and generating a statistical histogram for the video words; obtaining other video words in the space-time neighborhood of the video words according to the space-time context information in the space-time neighborhood of the video words and forming space-time video phrases by the video words and one of other video words which meets space-time constraint; clustering the space-time contexts in the space-time neighborhood of the video words to obtain context words and forming space-time video word groups by the video words and the context words; selecting the representative space-time video phrase from the space-time video phrases and selecting the representative space-time video word group from the space-time video word groups; and training a classifier by utilizing the result after one or more features in the video words, the representative space-time video phrase and the representative space-time video word group are fused.

Description

technical field [0001] The invention relates to the field of video analysis, in particular to a training method and recognition method for human action recognition. Background technique [0002] In recent years, with the rapid popularization and development of media such as film and television and the Internet, video has become the main tool for carrying information, and the amount of video data is growing explosively, and a large amount of new content will be generated every moment. In the face of massive video data, how to automatically obtain and analyze the information contained in it, and understand the actions, behaviors or events that occurred in it has become an urgent problem to be solved. [0003] Most videos record the activities of people who are the main body of social activities. How to enable computers to "see" the video or "understand" the actions of people in the video has become an important issue in computer vision, image processing, pattern recognition, m...

AI Technical Summary

Problems solved by technology

The problem with this method of detecting interest points is that sometimes the detected points are too few to capture all the features of human motion, which leads to a decline in recognition performance.

However, these early studies were limited to human action recognition in restricted scenes, such as specific perspectives, action figures, backgrounds, and lighting. In natural scenes, when the above-mentioned restrictions were removed, the performance of the method dropped sharply or even no longer applicable

Images