Downlink Human Detection Under Occlusion Based on Convolutional Neural Network

Abstract

As one of the hotspots in the field of target detection, pedestrian detection has very high value in the application fields such as driverless vehicle assistance system, intelligent monitoring system and service-oriented intelligent robot. The pedestrian occlusion studied in this paper can be divided into two types: human to human self occlusion and object to human occlusion. Aiming at the problems of missing detection, high false detection and low detection accuracy of small-size targets in real scene pedestrian detection methods, a pedestrian detection algorithm based on improved convolution neural depth network model is proposed in this paper. The algorithm improves the original SSD network model by extracting the lower level output feature map, and uses the abstract features of the output of different layers of convolutional neural network to detect pedestrian targets respectively. This method combines the multi-layer detection results and improves the detection performance of small target pedestrians. The experimental results show that the accuracy of the proposed algorithm is 93.8% on the INRIA test set, and the missed detection rate is as low as 7.49%.