Obstacle Detection System Based On SURF and Saliency-map

• Main Authors: Huang, Ya-Han, 黃雅涵
• Other Authors: Tao,Chin-Wang
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/7g427d

碩士 === 國立宜蘭大學 === 電機工程學系碩士班 === 103 === In this paper, we propose an obstacle detection system using SURF method, SVM model and saliency map. In addition, we present fuzzy weighting to adjust saliency map by histogram-based contract (HC) method. The proposed detection system composes of three major steps: First, the dense optical flow is used to extract the feature vectors of each pixel in the image sequence. The feature vectors are the training data of SVM. Second, the local feature points are detected by SURF method and are classified into the obstacle points and others by the trained SVM model in test stage. Finally, the obstacle points are combined with saliency map based on fuzzy weighting to find those with higher salient values. And then, these points are used to estimate the region of the obstacle. This method is a vision-based obstacle detection with single camera. It’s different from image subtraction method to detect the obstacle. Experimental results show that the system can accurately locate the obstacle position in the indoor and outdoor environments.