基于RGB-D多通道特征的行人检测
Pedestrian detection based on RGB-D multi-channel feature

针对行人检测易受物体遮挡以及光照变化干扰的问题，提出一种融合颜色与深度信息的多通道特征行人检测方法. 首先，颜色采用ChnFtrs方法中的通道，深度在其基础上引入法向量方向通道，并用快速图像特征金字塔来加速颜色和深度的通道特征的计算. 其次，通道特征作为级联AdaBoost的候选特征点集输入，分别训练得到颜色和深度分类器，按一定比例权重融合颜色和深度信息进行检测. 实验表明，该方法提高了检测精度，对光照变化、物体遮挡具有较好的鲁棒性.
As pedestrian detection is vulnerable to the interference of occlusion andillumination changes，we prorosed a novel pedestrian detection approach based on RGB-D multi-channel feature. First of all，RGB image adopt the channel that ChnFtrs used，while depth image introduce normal vector as channel.They both take fastimage feature pyramids to compute muti-channel feature.Then the cascade AdaBoostis trained with RGB and depth muti-channel feature respectively.Finally，we combine the color and depth information with certain weight proportion for last detect.The experiment shows that the proposed approach improves the detection accuracy，and has good robustness on illumination changes and occlusion