中文摘要：

为了快速检测行人，根据人体头顶特性快速找出头顶候选点，依此为基点利用边缘检测算法提取目标轮廓，将其作为待测区域;然后利用改进的颜色自相似特征、结合区域边缘直方图和一致模式LBP^特征，在Hough森林分类器上进行分类检测。实验结果表明，该算法在检测正确率及速率方面都得到了提高，对于复杂动态场景、遮挡现象以及目标自身形变具有较强的鲁棒性和较高的准确性，在多种行人数据集中测试取得良好的效果。

英文摘要：

In order to quickly detect pedestrians,we first rapidly found out the candidates of the top of the head according to the characteristics of human head tops,and according to this base point we extracted the object contour by using edge detection algorithm and took it as the area to be tested.Secondly,by using the improved colour self-similarity character and combining the region edge histogram and uniform pattern LBP26,1 characters,we conducted the classification and detection on Hough forest classifier.Experimental results showed that, the algorithm proposed in this paper gained the improvement in both the detection accuracy and speed,and had stronger robustness and higher accuracy against the cluttered dynamic background,occlusion and the deformation of the object itself.In many pedestrian datasets tests it achieved good results as well.