Research Of Video-based Moving Human Detection

Video based Moving human detection is a crucial issue in the field of computer vision and pattern recognition, and it is very important for image processing, video encode, intelligent surveillance and so on. A moving human detection system based on video sequence is constructed in this thesis, which includes two main issues: (a) moving object detection, extracting the moving region in the image, (b) and human detection, detecting the human in the moving region.For moving object detection, background subtraction based detection algorithms are mainly studied, and after analyzing several popular methods, a new algorithm based on color and contexture information is proposed: a single Gaussian model is used to model the statistical characteristic of gray value and Markov Random Field (MRF) is used to specify the contexture constraint.And for human detection, large numbers of positive and negative human samples are used to train a classifier off-line, which is used to detect the moving region on-line. The classifier used in the system is based on HoG feature in the form of cascade. By using the cascade classifier, non-human windows are rejected in the previous layers, and human windows can go through each layer successfully, which can accelerate the process and reduce the needed computational load to satisfy real-time applications.The experiments show that the proposed moving object detection method can fast detect the moving region completely, and it is robust against noises and disturbance of the background. HoG feature based human detection method can detect the static human in the image successfully. Finally, the moving human detection system works very well for detecting the moving human in the video sequence.