摘要
复杂图像中对特定目标的检测和定位是机器视觉领域的难点之一。提出使用中层视觉元素描述检测目标,以建立权值模板图像;然后对目标图像和权值模板进行加权SIFT特征匹配得到最优匹配位置,从而实现目标检测。该方法以自行车为检测目标进行实验,检测率达到86%,优于传统SIFT-Ada Boost和HOG-SVM检测方法。实验结果表明该方法能够减少复杂图像中背景干扰的问题,对于不同姿态的目标进行检测也有较强的鲁棒性。
Detection and localization of specific targets in complex images is one of the difficulties in machine vision. Uses the middle-level visual elements to describe the detection target to establish a weight template image; in order to achieve target detection, matches the weighted SIFT features of weight templates and the target images to get the optimal match position. Takes bicycles as the detection targets which have the detection rate of 83%, and our approach is better than SIFT-AdaBoost and HOG-SVM detection method. The result shows that this approach can reduce the background information interference in complex images, and has strong robustness to the targets with differ- ent poses.
