A Motion Feature Extraction Method Based on Vision

Abstract

The present invention relates to a vision-based motion feature extraction method, which specifically includes the following steps: step S1: extracting the motion vector of each pixel point based on continuous frames; step S2: extracting features with strong changes in pixel values ​​in the X, Y, and T directions Step S3: Constructing a cubic feature vector based on the direction-amplitude histogram of the motion vector with the feature point as the center; Step S4: Forming a coding vector for the local descriptor through a clustering algorithm. The spatiotemporal feature description based on the motion vector extracted by this method can better capture the motion information of the moving target, and the clustering recoding can form the lexicon frequency of the motion feature and improve the detection and recognition results of the spatiotemporal moving target.

Description

technical field [0001] The invention relates to the technical fields of computer vision and image processing, in particular to a vision-based motion feature extraction method. Background technique [0002] With the deepening of the application and promotion of the State Grid Digital Archives, the electronic archives have been standardized and used efficiently. New technologies and methods based on intelligent audio and video processing have been introduced into electronic archives management to solve the problems of large-scale storage, processing, and retrieval of video files. Intelligent video analysis is a very active field in computer vision, and the motion feature extraction technology of moving objects based on video is a hotspot in the research of intelligent video analysis. Intelligent video technology automatically analyzes video data without human intervention, extracts detailed information of targets, understands and analyzes targets or activity status, and reali...

Images