Research On Pedestrian Detection Based On Convolutional Neural Network

Pedestrian detection is one of the important research tasks in the field of computer vision.It has important research value and application prospects in areas such as automatic driving,safety prevention and control,human-computer interaction,and video surveillance.Pedestrian detection is mainly divided into two categories in the development process: One type is based on traditional machine learning methods,which mainly focus on manual feature extraction and feature classification;The other is a pedestrian detection method based on convolutional neural networks.This type of algorithm mainly focuses on the research of neural networks.The research content of this paper is to study the target detection algorithm based on deep learning in the application field.The main research content is as follows:First,the paper introduces the current domestic and foreign research and development of pedestrian detection and the difficulties of research technology,and then introduces in detail the traditional pedestrian detection algorithms: Histogram of Oriented Gradient(HOG)and Support Vector Machine(SVM),and conduct experimental detection of pedestrians in the campus.Secondly,two types of convolutional neural network models based on deep learning are studied.One type is convolutional neural network models based on candidate regions,mainly RCNN,Fast-RCNN,Faster-RCNN.The other is a convolutional neural network model based on image regression,mainly including YOLO and SSD.By analyzing the principle,structure and detection process of the two types of models,the Faster-RCNN,YOLO and SSD models are used to test the pedestrian database.By analyzing the detection results,it is proved that the two types of models can effectively detect pedestrians.Thirdly,in view of the difficulty of detecting small target pedestrians,based on the YOLOv3 network model,the integration of local and global features is achieved by adding a three-layer spatial pyramid pooling(SPP)module,which enriches the expressive ability of the feature map and improves it,the detection accuracy is further conducive to small target pedestrian detection.Finally,the improved YOLOv3 model and the original YOLOv3 network model are compared and tested.The accuracy and recall rates are selected as evaluation indicators,and the self-built campus small target pedestrian database is tested respectively.The comparison experiment results show that: after the improvement of this article,the model can effectively detect small target pedestrians,and the accuracy and recall rate are better than the original YOLOv3 model.