目标边缘特征增强检测算法

针对分组角点检测网络在目标检测过程中,由于目标尺寸过小或同类目标空间距离较小而导致检测失效的问题,提出一种边缘特征增强的CornerNet目标检测算法OEC。该算法通过分离特征的高低频信息提取更多的高频信息,增强目标的边缘轮廓特征,解决关键点定位不准确的问题,提高目标的框定效果,进一步提升检测精度。仿真结果表明,该算法对行人、车辆等目标检测效果均有提高,在COCO数据集上的检测结果与CornerNet相比,mAP提高0.9%,可应用于无人驾驶与智能机器人等场景。

Object Edge Feature Enhancement Detection Algorithm

In the process of CornerNet object detection, small objects and short spatial distances of similar object result in detection failure. Aiming at the problem, this paper proposes an object edge feature enhancement CornerNet object detection algorithm (OEC). The algorithm enhances object edge features through extracting high frequency information by separating the high and low frequency information of object features, which solves the problem of inaccurate location of key points to improve the object detection precision. After simulation tests, OEC has a good performance in detecting pedestrians, vehicles and other objects. The detection result on COCO (Common Objects in Context) dataset is 0.9% higher than CornerNet, which is able to apply to autonomous vehicle systems and smart robots.