Research On Identifying And Locating Urban Small Public Space Based On Deep Learning Processing Of Street View Images

Images processing,as an important research direction in the field of optical engineering,has ushered in new developments under the promotion and integration of deep learning,which can make the semantic information of images be better explored and expressed;the small public spaces hidden in every corner of the cities have not gotten extensive attention and efficient development due to their small area and inaccessibility.The development of these small public spaces can improve the efficiency of urban management planning and enhance the vitality of the city.Therefore,the identification and positioning of small public spaces is very important.The traditional small public space positioning research work requires a lot of manpower and material resources to carry out on-site investigation and research,which is costly and inefficient.Street view images have become an efficient and new data source due to their advantages of close-to-people perspective,comprehensive coverage of elements,and wide coverage.Street view images can express the contained space information of small public spaces in detail under the methods of deep learning and image processing.But traditional optical methods cannot express the semantic information of Street View images well.In order to solve the problem that the traditional small public space positioning research process is laborious and inefficient,and the traditional optical method cannot express the semantic information of the Street View image well,for the different color distribution and element components of different street view images,this paper studies a set of small public spaces identification and positioning that integrates three deep learning and image processing methods,image similarity matching,image object detection and recognition,and image semantic segmentation.The three methods complement each other.,complement each other,gradually deepen the processing of street view images and fully meet the characteristics of street view images of different degrees which can comprehensively and efficiently mine and analyze Beijing street view image information,accurately identify and locate small public spaces and has reference value and guiding significance for improving the urban research method and other urban small public spaces research work.The main research contents of this paper are as follows:(1)Baidu Street View images were obtained by using Python coding to access the Baidu API and the image distortion problem during street view image acquisition was solved by solving the distortion function.(2)The current research results are sorted out and combining with the characteristics of small public spaces,three methods of deep learning and optical image information processing are determined according to the color distribution and element components of street view images to identify and locate small public spaces.(3)For the case where the color distribution of street view images is obviously different,this paper adopts the image similarity matching method to find the street view images that are similar to the target image in the image set to be detected.image similarity matching method calculates the similarity of the pictures according to the perceptual hash algorithm and the cosine similarity algorithm,and outputs the street view images that reach the similarity threshold to complete the identification and positioning of the small public spaces.(4)For the case where it is necessary to find the street view images containing the target elements,this paper adopts the target detection and recognition method based on the YOLOv5 algorithm to identify the target elements in the street view images.The Labelme module is used to label the target elements and train them into a dataset that meets the detection requirements.The YOLOv5 algorithm is used to detect and identify the street view images containing the target elements and output the street view images that reach the set threshold confidence to realize the identification and positioning of small public spaces.(5)For the case where it is necessary to distinguish the relative area of the target elements in practice,this paper adopts the PSPNet semantic segmentation method based on the ADE20 K dataset to obtain the actual relative area of the target elements.Data enhancement and PSPNet semantic segmentation are performed on the street view images of the same collection vehicle and the same collection angle,and the depth pixel value of the target element is calculated,and the actual relative area of the target element can be distinguished to realize the identification and positioning of small public spaces.(6)The researched urban small public space identification and positioning based on street view image processing and deep learning is applied to Beixiaguan Street in Beijing,and the recognition and positioning of small public space is realized by processing the street view images of Beixiaguan Street.It can accurately identify and locate small public spaces according to the color characteristics and element distribution of street view images.