Pedestrians Detection Based on the Integration of Human Features and Kernel Density Estimation

Abstract

Kernel density estimation doesn’t need the distribution hypothesis of the background characteristics, and also it does not need to estimate the parameters. Therefore, it can deal with moving object detection under complicated background, but the algorithm application is limited by the selection of kernel function bandwidth. In view of this problem, the method of integrating the human features and kernel density estimation is presented in this paper, aiming at the pedestrian detection. First of all, selecting the kernel function bandwidth through the priori information of moving object, then, based on kernel density estimation, extracting the foreground (i.e. moving object), and finally, once again using the human characteristic to detect the video pedestrians. The experiments show, compared with the traditional methods, the method introducing the priori information can greatly reduce the computing burden of kernel density estimation. Even with the changes of the light and the outside noise interference, the method presented in the paper can accurately detect pedestrians and no-pedestrians. This method can be applied to the vehicle, animal detecting, but a priori information plays an important role.