The Research On Pedestrian Detection Technology Based On Deep Convolution Network

Pedestrian detection belongs to the category of target detection,which is the process of classifying and locating the specified objects in video frames or images.In real scenes,pedestrians have the characteristics of both rigid and flexible objects,and have greater randomness in posture and action.Additionally,the diversity of dress,the change of distance,and the different degree of occlusion,all lead to a large intra-class difference in the abstract characteristics of pedestrians.At the same time,in complex scenes,there are natural factors with strong randomness,such as light,weather and so on.The combination of these human factors and natural factors brings great challenges to pedestrian detection.This paper mainly studies pedestrian detection technology based on deep convolution network.On the basis of general detection technology,aiming at the characteristics of pedestrians in complex scenes,an efficient,reliable and targeted pedestrian detection algorithm is designed to effectively improve the detection accuracy under the condition of guaranteeing speed.The specific work and innovation of this paper are as follows:(1)Firstly,the background and significance of pedestrian detection technology are expounded,then the research status of pedestrian detection and related technologies including target detection and semantics segmentation at home and abroad are introduced.Then,the traditional pedestrian detection framework is simply described and the basic algorithm of deep learning is introduced.Finally,the evaluation index of pedestrian detection data and model is given.The standard,experimental platform and development environment are briefly introduced.(2)An efficient pedestrian detection method based on YOLOv2 is proposed.The model uses YOLOv2 as the basic detector.It is mainly designed for INRIA pedestrian detection data set.It can directly detect and recognize all targets in the whole picture at a time,and has a high efficiency.The whole model is designed and trained in an end-to-end manner,which can achieve higher detection accuracy at a higher detection speed.(3)A multi-scale perceptual pedestrian detection method based on semantic fusion technology is proposed.Faster-RCNN is used as the basic detector in this model.The model is mainly designed for Caltech pedestrian detection data set,which includes two parts: candidate region extraction network and classification network.The whole model is trained by CityPersons and Caltech data sets in two stages,and the detection accuracy reaches a high level.