Research Of Pedestrian Detection Based On Weighted Part Models

Pedestrian detection is widely used in the field of human-computer interaction, intelligent video surveillance and intelligent transportation. It is not only the basis for target classification and behavior to understand, but also one of the important topics in the field of computer vision research. However, due to the changes in complex background, posture and dress, and the existence of occlusion, pedestrian detection has become a more difficult problem. In this paper, based on the deformable part models, the pedestrian detection method for crowded scenes is studied by using statistical analysis and machine learning theory.Firstly, aiming at the problem that the deformable part model detector starts to fail in crowded scenes, a pedestrian detection method based on weighted part models is proposed by analyzing the features of pedestrians in crowded scenes and characteristics of the deformable part models. The weighted part models are assigned different weights for different parts. And the weights are trained by latent support vector machine.Secondly, aiming at the search space of the existing pedestrian detection method is too large, resulting in the reasoning process is slow and low accuracy. A pedestrian cascaded detection algorithm based on selective search segmentation is proposed according to analysis of the selective search segmentation and cascade detection. In this method, image reprocessing based on selective search segmentation is done, leaving less effective windows which contains more objects, on this basis, the cascade detection is used during the process for pedestrian detection.Finally, by comparing with existing methods, the feasibility and effectiveness of pedestrian detection method is verified from two points of view: model and detection method.