Abstract
Pedestrian detection as a basic identification technology, providing technical support for many areas such as security monitoring and automatic driving, and has a wide range of application scenarios. This article is based on the Single Shot Detector(SSD)target detection algorithm, a pedestrian detection system based on the SSD target detection framework was trained using a self-built occlusion pedestrian data set for the specific goal of occluding pedestrians, then the features learned by deep convolution neural network are more robust. For the occlusion pedestrian detection, the network structure of the SSD model has been modified accordingly. The depth can be divided into convolutions to simplify the feature extraction network, reduce the number of network parameters, reduce the training difficulty, and improve the algorithm operation speed. Further, in the prenetwork of the SSD model, the SE-Inception structure was added, and the a priori frame in the network was redesigned to make it easier to match the shape of the pedestrian. In the case of pedestrians who are prone to environmental occlusion and mutual occlusion, data on occlusion pedestrians is added to the training set of the model. The Repulsion Loss enhanced model is used to block the detection of pedestrians, and the experimental results show that the detection performance of the improved SSD model has been greatly improved.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call