Stereo Vision-Based Front Vehicle Detection and Its Distance Estimation System

• Main Authors: WENG, KAI-WEN, 翁凱文
• Other Authors: WU, CHIEN-CHUNG
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/4urzsz

碩士 === 南臺科技大學 === 資訊工程系 === 104 === As the rapid development of car industry and gradual complexities of traffic flow, casualties caused by car accidents has increased as well. To eliminate car accident, car use active safety and passive safety related product has been development constantly in car industry. This research uses video sensor to detect vehicle and its location and assists drivers in driving more safely through the alerting system. In this paper, the testing video is recorded through presetting two GoPro Hero 3+ to set up three-dimensional dual camera which was installed in front of the drivers’ seat in order to record three-dimensional car movement video. The testing figures is obtained from analyzing and processing these videos. In terms of vehicle detection, first of all, Haar feature is used along with Adaboost algorithm to detect the vehicle area in the videos. After calculating the feature dimension through HOG-PCA, uses SVM algorithm to identify vehicle. Meanwhile, in vehicle detection, particle filter is added to HOG-PCA feature and SVM classifier. Vehicle detection can keep the weight of particles. Through this modification, the accuracy for vehicle detection while cars driving through dark lanes can be improved. Apart from detecting the locations of vehicle, this research find out vehicle features through SURF algorithm, compares the features of two images on the left and right through SGM, and calculates image depth information based on SGM features. The distant figures can be obtained through known camera parameters conversion. This information is beneficial to drivers to clearly understand the distance between each car.