Human detection in semi-dense scenes using HOG descriptor and mixture of SVMs

Human detection has recently received significant attention in the field of computer vision. Accurate detection of human bodies is an essential component required by a variety of applications such as automated surveillance, advanced user interface and sport analysis. In this paper, we present a new method for human detection in video frames, using learning algorithm, feature extraction and sliding window. People detection is a challenging topic, especially in complex scenes when occlusion occurs. In this paper a new technique based on the body part detection and HOG (Histogram of Gradient) features is proposed for human detection with occlusions. Proposed HOG descriptor provides lower feature vector length in comparison with state of the art methods approaches. Using human detectors alone, cause to high frequent false positive. Thus, we propose a novel classifier-fusion learning algorithm, instead of single classifier. The experimental results show that the new proposed method provides lower false positive and raise higher precision and recall in comparison with state of the art methods.