The object detection algorithm by gradient and curvature information HOG descriptor

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 more distinguished feature descriptor, 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.