行人检测是计算机视觉技术中一个热门的研究热点, 在汽车辅助驾驶和视频监控等方面具有重要作用. 由于传统的可变形部件模型(deformable part model, DPM)采用滑动窗口检测方式, 在背景区域花费大量检测时间会导致检测速度降低, 因此提出了一种基于 BING-casDPM 的快速行人检测算法. 首先基于二进制化梯度范数特征(binarized normed gradient, BING)训练一个二级支持向量机(support vector machine, SVM)分类器, 通过该分类器快速标定出所测图像中包含各类物体的候选区域; 然后根据候选区域窗口的特点进一步提取待检测框; 最后将待检测框作为输入, 使用级联 DPM(cascade DPM, casDPM)模型进行精确检测, 并将结果返回至原图. 实验结果表明, 该算法在基本不降低检测率的情况下, 其检测速度比经典 DPM 模型检测速度提高了约 16 倍, 比 casDPM 模型提高了约40%.

Pedestrian detection is a hot research field in computer vision technology, and it has important applications in automobile assisted driving and video surveillance. In view of the fact that traditional deformable part model (DPM) uses a sliding window for pedestrian detection and is time-consuming in the background area, a fast pedestrian detection algorithm based on BING-casDPM is proposed. Firstly, a cascaded support vector machine (SVM) classifier is trained based on the binarized normed gradients (BING) feature. The candidate areas containing all kinds of objects as seen on the test image are quickly calibrated through the classifier. Then, the detected box is extracted according to the characteristics of candidate windows, and finally pedestrians are accurately detected from the detected box using casDPM model with the result being returned to the original image. The experiment results show that the algorithm proposed in this paper is about 16 times faster than the classical DPM model, and about 40% faster than the casDPM model in terms of the speed of detection when detection rate remains stable.