Research And Implementation Of Pedestrian Detection And Attribute Recognition Based On DenseCDNet And Multi-task Deep Learning

Pedestrian detection and attribute recognition are important research contents of intelligent video processing,and have broad application prospects.If we can effectively detect and recognize the human feature image contained in the image or video,the recognition cost can be greatly reduced.Therefore,this paper studies and implements pedestrian detection and attribute recognition algorithm based on convolution + deconvolution structure and multi-task deep learning.By improving the deep learning benchmark network,this paper combines the advantages of Dense Net and YOLOv3 in pedestrian detection,and proposes Dense?YOLO in the aspect of pedestrian attribute recognition,Dense CDNet algorithm based on convolution +deconvolution structure is proposed for attribute recognition,which improves the model effect.The main research work of this paper is as follows:(1)In order to improve the performance of pedestrian detection model,Dense Net is integrated with YOLO framework,and dense is proposed Dense?YOLO network structure: taking YOLOv3 as an example,the double backbone structure is introduced into the network structure design of YOLOv3,and the features of multi-level connected backbone networks are integrated,so as to realize the diversification of feature graph types and extract more abundant network features.The experimental results show that the proposed network structure fusion model improves the detection accuracy of the original benchmark network,and the convergence speed is faster,which can effectively detect the whole pedestrian and head.The idea of heterogeneous network feature fusion used in this paper can be widely used in deep learning feature fusion algorithm.(2)In this paper,a basic block module of deep learning network,convolution +deconvolution structure,is proposed.When it is used for feature space conversion and feature extraction,it can comprehensively utilize the information of other points around the feature points,so as to improve the classification accuracy of neural network.The convolution + deconvolution structure proposed in this paper is applied to Dense Net and Res Net,and the accuracy is improved.(3)On the basis of the above research,a pedestrian attribute recognition method based on Dense CDNet multi-task branch network based on convolution +deconvolution structure is proposed.Compared with the existing methods,this method can extract multi-attribute features by setting multi branches,and can achieve good attribute recognition effect without pre-training the attribute feature recognition module.