Abstract
A method for searching the moving object by the color stereo camera fixed on a moving platform is proposed in this study. SIFT corner detector is used to match points in the left and right images. In addition, additional matching points are obtained by the parallax concept, control points, color segmentation and epipolar line geometry principle. Moreover, the continuous matching points are calculated by depth information, platform information and searching window concept. Finally, the moving object position is obtained by the compensation of the background using an appropriate transform function. Different from others, the color segmentations with its own depth for each in an image is used to improve the correct rate of transformation. Besides, the corner detector and the parallax concept are used to solve smooth area without enough characteristic points and the best transfer function is determined by multi-RANSAC. The percentage of correct matching pixels approaches to 92% by multi-RANSAC. The current method can be used for searching either outdoor or indoor moving objects and the results show that two persons outdoors can be found and located successfully with this proposed algorithm. This research develops a moving object detection technique to improve the object recognition. © 2012 American Scientific Publishers. All rights reserved.