Human Detection Based On Feature Extraction

Intelligent video surveillance can comprehend monitor videoautomatically using computer technology, realizing object detection,recognition, tracking and alerting. It is always the focus both in academic orindustrial area since it was proposed. Intelligent video surveillance is a branchof Computer Vision. Computer Vision is an interdisciplinary subject of imageprocessing, machine learning, pattern recognition, and artificial intelligence.Human recognition is the frontier direction and research focus in intelligentvideo surveillance area. But current human detection technology has somefaulty, eg, slow running speed, low precision, high missed rate. So humandetection technology is to improve detection speed while guaranteeing thedetection accuracy.At present, the popular technique is human detection based on machinelearning. This paper give a solution of Real-Time human detectionsystem(RTHD). RTHD classifies human body using cascade classifier based onHOG(Histogram Of Gradient) feature. Cascade classifier is trained byadaboost algorithm. Motion region in the monitor video can be classified usingit.RTHD consists of4modules:(1). Motion region detection module.(2).HOG feature extraction module.(3). Cascade classifier training module.(4).Human detection module.This paper uses3-temporal difference algorithm to get motion region, anduses integral image to get HOG feature of training samples. Cascade classifieris trained with discrete adaboost algorithm. We use cascade classifier to checkwhether motion region is human body. Experiments show that motion region detection algorithm used by ourpaper can get complete foreground area quickly, and adapt to the change oflight. More over our system could inhibit interruption by noise with the help ofprinciple of noise reduction. Cascade classifier based on HOG feature canjudge whether motion region is human body precisely and quickly.