Remote Pedestrian Detection Algorithm Based on Edge Information Input CNN

Abstract

In order to solve remote pedestrian detection problem, the target need to be detected in the absence of information, a new pedestrian detection algorithm based on Convolution Neural Network (CNN) is proposed. The algorithm uses shallow layer edge features combined with grayscale images to replace the RGB color information of the original image, as an input to the Convolutional Neural Network to increase the amount of effective information. Then, in deep learning training process, the cross entropy is combined with the learning rate to optimize the cross entropy function. Finally, the improved Convolutional Neural Network is trained on four common pedestrian hybrid datasets to apply it to the remote pedestrian intrusion detection of the railway industry using transfer learning. The experimental results show that compared with the existing Convolutional Neural Network remote pedestrian detection algorithm, the new method can effectively improve the accuracy of detection 2% and has a good universality.