Deep learning based pedestrian detection

Abstract

The application of pedestrian detection is extremely extensive, such as in the field of intelligent surveillance, unmanned vehicle, and pedestrian analysis. With the development of deep learning, deep learning based pedestrian detection method has greatly improved the accuracy of pedestrian detection; however, the computation cost is very large. In this paper, a pedestrian detection framework based on deep learning is proposed. An optimized PVANet is first utilized to generate the feature maps; then, the region proposal network is utilized to generate the pedestrian candidates and the corresponding scores; finally, the enhanced decision trees is utilized to complete the issue of pedestrian detection. Experimental results conducted on Caltech pedestrian detection dataset demonstrate the effectiveness of our method.