Traffic Volume Detection And Analysis Based On Remote Sensing Image Processing

Recently,with the development of transportation,the scale of the road network is expanding,and the requirements for traffic management are becoming higher and higher.Traffic volume data is the basis of traffic management.Traditional traffic volume acquisition methods have some shortcomings,such as high cost,complex installation,deployment and maintenance,and small coverage area.With the improvement of remote sensing image resolution,it contains more features,and remote sensing image has the advantages of low cost and wide coverage,which provides a new way for traffic volume detection.However,there are some problems to be solved urgently in the current remote sensing image processing methods.Firstly,the current road extraction methods are inefficient and time-consuming,and the regions with similar characteristics are not easily distinguished from the roads,which leads to false alarm.Secondly,the traditional methods depend on the choice of the best threshold.For remote sensing images from different sources,the accuracy of the vehicle detection algorithm is low,resulting in large errors in the calculation results of traffic indicators.Finally,the automation of speed estimation method is not high,and manual assistance is needed in image processing,which is time-consuming and laborious.To solve the above problems,this paper proposes a traffic detection and analysis method based on remote sensing image processing.This paper aims to automatically extract traffic volume data from remote sensing images.Superpixel word bag is constructed by superpixel segmentation and feature extraction method,and image feature vectors are extracted to train support vector machine to extract road area.Vehicle detection in image is based on deep learning method,combined with open data sets.Based on image matching,vehicle template is matched between panchromatic image and multispectral image.In addition,this paper calculates the displacement and speed of vehicles,and carried out traffic volume analysis.The main contents of this paper are as follows:(1)A framework for traffic detection and analysis based on remote sensing image processing is proposed.This paper presents a framework of traffic detection and analysis based on remote sensing image processing,aiming at automatic traffic extraction and analysis.The framework takes remote sensing image processing as the core,proceeds from road extraction,vehicle detection and traffic flow calculation and analysis,and combines traditional image processing and deep learning methods to construct a set of automatic road traffic extraction method flow.(2)Road area extraction based on feature extractionThis paper presents a road extraction method based on feature extraction.Remote sensing image covers a wide area and contains complex features,so it is necessary to extract road area efficiently and accurately.In this paper,we use the superpixel segmentation method,combined with the idea of superpixel bag of words,and train the support vector machine with the texture and color features to extract the road area in the image efficiently.(3)Vehicle detection Based on Deep LearningThis paper presents a vehicle detection method based on deep learning.Vehicle target is the focus of traffic volume detection,so the accuracy of vehicle detection is required.In this paper,the results of road area extraction are used as input,and candidate regions that are classified by convolution neural network are generated based on selective search algorithm.The accuracy of vehicle detection algorithm has been effectively improved.(4)Traffic volume calculation and analysis based on image matchingThis paper presents a traffic calculation and analysis method based on image matching.The vehicle template is determined by the vehicle position,and matched with the target image of multispectral image.The matching efficiency is improved by predicting moving blocks,and the similarity matrix is obtained to calculate the displacement and speed of the vehicle.(5)Design and implement the prototype of traffic volume detection and analysis systemBased on the above methods,this paper designs and implements the prototype system.This paper introduces the overall architecture of the prototype system,and elaborates the detailed design of the system in layers.The implementation process and prototype examples of the system are introduced and compared with similar methods.