Research And Implementation Of Pedestrain Attribute Recognition Technology Based On Semantic Segmentation Research And Implementation Of Pedestrain

A smart city is a new term given to society by technology,and a ca-mera is an important infrastructure for building a smart city.How to use camera information efficiently and effectively play an important role in people's daily life and maintain social order.Pedestrian information ac-counts for a large proportion of camera information,Pedestrian search,pedestrian recognition and pedestrian re-recognition can be performed through pedestrian information in the camera.Therefore,the study of pe-destrian information has a strong practical significance.This paper re-cognizes the pedestrian attribute information of nature scenes.We sum-maries the previous research work on pedestrian attributes and conclude that previous work mainly studies the existence of attribute information.However,the position of pedestrian attribute takes an important part in some areas,like pedestrian re-identification.So,this paper focus on the acquisition of the position information of pedestrians.This paper uses semantic segmentation to get the position infor-mation of pedestrians.In the face of the current situation of no available pedestrian attribute semantic segmentation data set,this paper makes a pedestrian attribute semantic segmentation data set.Based on this data set,we use different semantic segmentation frameworks for experiments.We use simple semantic segmentation structure for experimental veryfication and train model that is higher precision and more complex to obtain the miou reference value of this data set.Besides,we analyze the correlation between miou and the three aspects of attribute category,accuracy and image clarity,and concludes that miou is positively related to the number of attributes,accuracy and image clarity.Finally,we modify the semantic segmentation model from two aspects:reducing the complexity of the model and making better use of features.Some suggestions for related research are provided.