Pedestrian Detection Based On Fast R-CNN and Batch Normalization

Zhong-Qiu Zhao, Haiman Bian, Donghui Hu, Herve Glotin

Abstract: Most of the pedestrian detection methods are based on hand-crafted features which produce low accuracy on complex scenes. With the development of deep learning method, pedestrian detection has achieved great success. In this paper, we take advantage of a convolutional neural network which is based on Fast R-CNN framework to extract robust pedestrian features for efficient and effective pedestrian detection in complicated environments. We use the EdgeBoxes algorithm to generate effective region proposals from an image, as the quality of extracted region proposals can greatly affect the detection performance. In order to reduce the training time and to improve the generalization performance, we add a batch normalization layer between the convolutional layer and the activation function layer. Experiments show that the proposed method achieves satisfactory performance on the INRIA and ETH datasets.

Conflicts: google.com