曲率-HOG目标检测算法研究

Research of Curvature-HOG Object Detection Algorithm

HOG算法充分利用梯度方向与幅值分布信息,在行人、车辆等特定目标检测定位中得到广泛应用,针对如何在HOG算法中融合梯度信息和曲率信息,本文提出基于HOG的梯度曲率相结合的目标检测算法并将其应用于目标检测问题中。算法首先计算图像的曲率信息和梯度信息,利用梯度的幅值和方向作为约束条件,统计曲率的分布直方图;然后将曲率直方图特征与HOG特征相连接,构成新的性能更好的特征描述子,最后将这一特征描述子用于检测实验。在ETZH形状数据库和INRIA马匹数据库上的实验结果表明,本文算法能够更好地检测目标并获得较高的精度。

The HOG descriptor takes full advantage of the information of gradient direction and amplitude distribution,which is widely applied in pedestrians,vehicles and other specific object detection and location task.In order to introduce curvature information and gradient information to the HOG descriptor,this paper proposes an improved object detection algorithm based on HOG algorithm.Firstly,calculate the curvature information and gradient information,apply the gradient amplitude and direction as constraint condition to obtain curvature distribution histogram,and then build a new feature descriptor with better detection performance,finally,use the new feature vector to detect objects.Experiment results in the shape ETZH database and INRIA horse database show that the proposed algorithm can obtain better detection performance and a higher accuracy.