近年来具有代表性的工作之一是Felzenszwalb等人提出的可变形部件模型。文中从可变形部件模型中存在的一些问题展开讨论,并提出了一种基于动态加权可变形部件模型的行人检测算法。在各个部件的检测中使用动态调整权值的方法获得更加准确的判断,从而识别处于复杂环境中的行人。实验结果表明,该方法能够有效识别复杂环境中传统DPM方法难以识别的行人。
Pedestrian detection is an active research in computer vision. One of the representative works in pe- destrian detection in recent years is the deformable part model (DPM). In this paper we discuss the disadvantages in DPM, and propose a new method based on the dynamic weight adjustment in DPM. The weight of the parts in DPM is adjusted dynamically to obtain more reliable results for identifying pedestrians in complex environments. Experi- ments show that this method can identify pedestrian more effective than the traditional DPM in complex environments.