基于优化HOG特征计算的非完整人体特征检测

Incomplete Human Feature Detection Based on Optimized HOG Featur Computation

摘要:

针对灾难中被困者的识别问题，提出了一种遮挡状态下的非完整人体特征检测方法。基于头部、腿部等可能暴露在外的人体特征，先构造相交检测窗口，重新检测目标图像块内的方向梯度直方图（HOG）特征，并结合基于子单元插值的方法计算块内特征，从而实现了基于优化HOG特征的非完整人体特征检测。实验结果表明，该优化HOG特征计算后的非完整人体检测方法可显著提高人体检测的检测速度和准确性，降低误检率。

Abstract:

A method of incomplete human feature detection under occlusion is proposed to identify trapped persons in disasters. In light of exposure of human characteristics such as the head and legs, the intersection detection window is constructed, and the features of histogram of oriented gradient ( HOG ) within the targeted image block are redetected. The intra-block features are calculated by combining the sub-unit interpolation method, so that the incomplete human feature detection is feasible on the basis of optimized HOG features. Experimental results show that the optimized human body detection method based on HOG feature calculation can significantly improve the detection speed and accuracy of human detection, and reduce the false detection rate.