Vehicle Information Extraction And Retrieval Based On Video

Video-based vehicle information extraction and retrieval technology in the field of contemporary traffic monitoring has a very important significance in the intelligent transportation,digital city and so on.Based on video collected from traffic camera,this paper mainly studies the detection of vehicle,vehicle information extraction,vehicle retrieval,data query and so on,which mainly including the following aspects:(1)Research on vehicle detection technology.Because the background of the video is complicated and changeable,this paper presents a method of improving the background extraction method,and update constantly,so the background is accurate and stable.This paper use contour detection method,the vehicle contour is extracted by background difference,chain code segmentation and convex hull optimization is used,then the vehicle is detected.As this article uses the global detection method,we can detect car traveling in different directions,the adaptability is stronger.(2)Research on vehicle information extraction technology.In this paper,we do research on color extraction and vehicle recognition technology,obtain the low frequency region containing the color information by Haar transform,convert the RGB space into HSI color space,extract the color histogram,and then use BP neural network to obtain the vehicle color.According to the attributes such as the perimeter and area of the vehicle extracted from the video,the SVM classifier is used to identify the vehicle in combination with the HOG feature.(3)Research on Vehicle retrieval algorithm.This paper studies image retrieval based on image feature,as SIFT is not symmetric invariant,we propose an improved SIFT feature ISIFT,which preserves the advantages of all SIFT features while preserving symmetric invariance.At the same time,we discuss the methods for reducing mismatched points.In order to solve the problem of slow retrieval,this paper studies the pHash feature and combines it with ISIFT to improve the search speed in some cases.(4)Software design and implementation.The vehicle information extraction and management software is designed and realized with above research algorithm,it realizes vehicle detection,information extraction,database management and image retrieval based on video collected from traffic openings.Finally,we use several videos to test the function and performance of the software.