一种基于深度信息的人头检测方法

A Head Detection Method Based on Depth Information

针对目前人头检测方法对光线变化敏感和易受阴影干扰的问题,提出了一种基于深度图像的人头检测方法。首先通过运动目标检测,得到运动人员所在区域;然后对该区域使用改进的立体匹配算法,该匹配算法对传统的WTA匹配算法进行改进,只对强纹理点进行匹配,对弱纹理点只进行视差验证,并根据三角投影原理计算出深度图。由于深度图中人员与周围场景的深度分布不同,根据深度分布将人头区域提取出来,得到候选区域,最后将候选区域经过形态学运算并根据区域轮廓的特征来判断是否为人头。实验结果表明:该方法在不同光线环境条件下的检测正确率为94%以上,误检测率仅为5.77%,检测精度高,对光线和阴影的抗干扰性良好,能够很好地适应复杂环境。

For the problem head detecting method is sensitive to changing light and vulnerable of shadow interference, a head detection method based on the depth map is proposed. Firstly, get moving target region by the moving target detection. Then a improved stereo matching algorithm is used to the target area, a matching algorithm based on the traditional WTA matching algorithm, we just need to match strong texture point by stereo matching algorithm, verify disparity of weak texture point and we can calculate the depth map. Because the depth distribution between people and surrounding scene is different in the depth map, we can extract the head region depend on the depth distribution, and then get candidate region through expansion and corrosion operation. Determine whether the region is the head with the characteristic of the head. Experimental results show that the method tested under different environmental conditions is 95.69%,error detection rate is 5.77%,high precision,strong anti-interference performance to light and shadow and ap-plicable to complex environment.