Pedestrian Detection Using 19-Layer Deep Convolution Neural Network

Abstract

Pedestrian detection is one of the key technologies in automotive safety, robotic and intelligent video surveillance. Recently, deep convolutional neural networks have achieved significant effect in image classification and retrieval tasks. In this paper, we propose a novel deep convolutional neural networks model for pedestrian detection to simultaneously extract and classify pedestrian features. The proposed model is a 19 layers network which consists of 7 convolution layers, 3 pooling layers, 6 relu layers, 2 normalization layers and a classification layer. The classical back propagation algorithm is adopted to train the model based on a self-build pedestrian datasets. Then, we test the performance of the proposed deep convolutional neural networks model on public INRIA, CVC and TUD pedestrian datasets, which achieves the detection accuracies of 87.52%, 91.98% and 89.98% respectively and outperforms the state-of-the-art conventional methods.