GoogLeNet-like Model for Pedestrian Attribute Detection in Surveillance Environment

Social public security has attracted worldwide attention in these years. Automatic pedestrian attribute detection is of great significance to improve monitoring efficiency. Low resolution of surveillance images and multi-angle problem of image acquisition are the main challenges of current pattern recognition researches. In order to improve the pedestrian attribute detection accuracy, we proposed a GoogLeNet-like model with inception structure which replaced the optimal local construction by a series of dense substructures. The experimental results show that the proposed method outperforms the previous works.