A Study Of Vehicle Detection And Distance Measurement In Driving Assistance System Based On Machine Vision

Driving assistance system based on machine vision is the premise of developing intelligent vehicles, improving driving safety and reducing traffic accidents. There are two hot spots of research in this field--one is accurate real-time detection and recognition technology of front moving vehicle, the other is measurement algorithm of the distance between front moving vehicle and the vehicle behind. As is well known, compared with ultrasonic, laser, and radar etc.,detection and recognition technology based on machine vision boasts advantages such as more information, more in line with the habit of human eye in capturing information, lower cost etc..Meanwhile, in order to improve the practical application value, the algorithm must be highly real-time, robust and accurate. On the basis of previous vehicle detection technology, this thesis makes new attempt on the development of vehicle detection technology. Also, theoretical derivation is made for the front moving vehicle distance measurement algorithm based on machine vision. To sum up, the followings are the tasks finished in this thesis:1. The thesis makes a detailed explanation of purpose and significance of the study, i.e., vehicle detection and distance measurement in driving assistance system based on machine vision; and then makes an introduction of current researches in this field.2. The thesis makes a detailed introduction of several categories of image feature extraction algorithm which are commonly seen and frequently used, and explains the learning methods of two kinds of machines-- AdaBoost and SVM. Meanwhile, feature extraction algorithm adopted in the thesis is determined, and machine learning classifier is opted.3. The thesis has proposed that real time accurate detection and recognition for front moving vehicle can be achieved with the combination of utilizing edge features of vehicle and Bag-of-Features(BoF) model. Detailed introduction of two parts is made-- generationof vehicle hypothesis existence area and validation of hypothesis area. In this algorithm,hypothesis existing area of vehicle is obtained by using Sobel edge detection processing generation after pretreatment of images; then, K nearest neighbor algorithm of Bag-of-Features is used to verify the hypothesis existing area, which can improve the accuracy of algorithm detection.4. Principles of camera imaging is studied, and formula for distance-measuring is analyzed and deduced. Zhang-Zhengyou’s camera calibration method is adopted to obtain internal parameter, which is then brought into formula for distance-measuring, and measurement result with relatively high accuracy is achieved.