Target Range Based On Visual And Infrared Automotive Night Vision System

Intelligent Transportation Systems (IST) is becoming the world's research focus in automotive safety areas as the solution to the problem of road traffic safety. Automotive night vision system is very important to safe driving at night, which can make the driver have a better view to observe the distant obstacles and improve driving safety at night, so automotive night-vision technology has been rapid development. Target ranging system, which is the key to improve drive safety and reduce traffic accidents, is an important component of the Intelligent Transportation visual navigation system; In addition, accurately measure target distance is also the premise of automatic or auxiliary driving.This paper study how to get the target ranging based on binocular stereo vision by video and IR image. The target ranging system includes the system model structure, software simulation platform and systems ranging programs. This thesis discusses the following four parts: image pre-processing, camera calibration, image matching and the target distance.(1) Image pre-processing section describes the common image pre-processing algorithms, including image filtering, image enhancement, morphological processing and so on. In this paper, using multi-scale Retinex algorithm to enhance the low visibility of infrared image, and then propose an improved image enhancement algorithm based on S curve Multi-Scale Retinex. The results show that the method can enhance image contrast and make the target clearer, is conducive to observe the moving target.(2) Camera calibration section discusses the camera imaging model and camera calibration method, taking into account the experimental environment and laboratory equipment, the plane template calibration method is used which has the calibration of high precision and convenient to operate.(3) Image matching section describes the four key elements of image matching and the classification image matching algorithm.This thesis uses image matching algorithm based on the SIFT feature descriptor,and then analyzes SIFT similarity measure and the image mismatch. By selecting the appropriate ratio threshold and using Euclidean distance for SIFT feature vector matching, the result is not very satisfactory.Then using RANSAC algorithm to eliminate mismatch points and improve the correct match rate.(4) Target ranging section introduces the structural model of binocular stereo vision, and then describes two situations: under the conditions of parallel binocular stereo vision the triangulation principle depth measurement and under the conditions of any location binocular stereo vision depth extraction measurement. Finally,capture an image from the scene by two camera and calculate the target distance,then compare with the actual distance and analysis error.