Classification And Recognition Of Pedestrian Attributes Based On Deep Learning In Surveillance Scenarios

As the number of cameras in cities has increased significantly,this provides great convenience for accident and pedestrian tracking and surveillance.Due to the surveillance information of the camera,the effective information is less.If the manual search for a particular pedestrian,the workload will be huge.The study is based on the method of deep learning to classify and identify the attributes of pedestrians.In this way,we can quickly identify the specific pedestrian properties,save the target recognition time,improve the accuracy of classification.In a real video surveillance scenario,visualized pedestrian properties,such as gender,backpacks,and type of clothes,are important for pedestrian retrieval and pedestrian re-identification.There are three main problems in the classification and identification of pedestrian: 1)The attribute class of the data set is not detailed enough;2)Pedestrian attribute recognition time is not real time;3)The relationship among pedestrian attributes is ignored.In order to solve the above three difficulties,we propose three different methods.First of all,we use of web crawler technology to build their own data sets,mainly focusing on the clothing type and head features of the fine-grained classification.Then,the pedestrian attribute is trained and tested by Faster R-CNN based algorithm,and the attribute recognition time is explored.Then,for the problem that the relationship between attributes is ignored,a fusion strategy is proposed.We merge DeepSAR and DeepMAR models,perform multi-attribute learning on pedestrians.The feature maps are generated in the training procedure.We vectorize the attributes of pedestrians,make weighting operations on multiple vectors,and then input to the loss layer for iterative training.Through the above three solutions,combining with the experimental results,we make a fine-grained classification on pedestrian attribute data set.We demonstrate that the improved method of Faster R-CNN can shorten the attribute recognition time.Considering the pedestrian attribute context,pedestrian attribute recognition performance has a better improvement.