Robust Pedestrian Detection Based on Parallel Channel Cascade Network

Abstract

Promoted by Smart City, pedestrian detection under wide-angle surveillance has attracted much attention. Aiming to the small-size pedestrians have poor resolution and different degrees of distortion in visual picture from wide-angle field of view, a robust pedestrian detection algorithm based on parallel channel cascade network is proposed. The algorithm, an improved Faster R-CNN (Faster Region Convolutional Neural Networks), first obtains the differential graph and original graph to construct parallel input, and then introduces a new feature extraction network, which called the Channel Cascade Network (CCN), further designs parallel CCN for fusing more abundant image features. Finally, in Region Proposal Network, the size distribution of pedestrians in the picture is counted by clustering to best fit the pedestrian date sets. Compared with the standard Faster-RCNN and the FPN, the proposed algorithm is more conducive to the small-size pedestrian detection in the case of wide angle field distortion.