Human behavior recognition method based on feature optimization and multi-feature fusion

Abstract

The invention discloses a human behavior recognition method based on feature optimization and multi-feature fusion, which comprises the following steps: 1, reading a video frame image and carrying out de-noising preprocessing, and carrying out de-noising processing on an image sequence by adopting a Gaussian filter to enhance the image quality; 2, HOG features and SURF features being extracted from the preprocessed video frame images respectively; 3, performing dimension reduction processing on the HOG features by using a PCA algorithm; 4, carrying out secondary dimension reduction processing on the HOG features after PCA dimension reduction by applying a Pearson correlation coefficient, a Spearman correlation coefficient and a Kendall coefficient; and 5, carrying out feature fusion on the SURF feature vector and the HOG feature vector after secondary dimension reduction, and carrying out classification and identification by using a support vector machine. Through the secondary dimensionality reduction and feature fusion technology, redundant features are removed, the calculation complexity is reduced, the recognition accuracy is improved, the method has the advantage of exceeding a generative model, and in addition, the method is high in robustness for noise and other influence factors and has good practicability.

Description

technical field [0001] The invention relates to the fields of image processing, video processing, pattern recognition, etc., and specifically relates to a human behavior recognition method based on feature optimization and fusion of multiple features. Background technique [0002] Human behavior recognition has broad application prospects, such as intelligent video surveillance, video summarization, intelligent interface, human-computer interaction, sports video analysis, video retrieval, etc. Generally, action recognition involves two important issues, one is how to extract useful motion information from raw video data, and the other is how to establish a motion reference model so that training and recognition methods can effectively deal with classes with varying spatial and temporal scales. similar behavior within. Behavior recognition can use various factors, such as human posture, optical flow, motion trajectory or contour, spatiotemporal features, etc. In recent years...

AI Technical Summary

Problems solved by technology

A single feature is often affected by factors such as human appearance, environment, camera settings, etc., and cannot accurately and comprehensively describe human motion, thus limiting the improvement of behavior recognition accuracy

Dalai et al. proposed an algorithm that fuses HOG features and LBP features that represent structural information in local areas of images, which can effectively improve the recognition rate, but at the same time, there are problems such as too large HOG feature dimensions, high complexity, and long time consumption.

Images