摘要： 针对夜间动态背景下的行人检测中分割算法受光照条件影响大、误识别多等问题，提出双阈值分割算法和以多目标跟踪为核心的算法框架。新的分割算法解决了行人亮度分布不均时的分割问题，同时在新的框架下可以综合多帧的处理结果进行综合判断，通过将基于支持向量机的识别算法和多目标跟踪算法的融合，降低了系统的计算量，且比一般的系统具有更高的识别率。

关键词: 行人检测, 多目标跟踪, 支持向量机

Abstract: This paper proposes a dual threshold segmentation algorithm and a multiple-object-tracking-based framework for the problems of nighttime pedestrian detection in dynamic scenes, such as segmentation greatly effected by illumination and high false detection rate. The segmentation method performs well even if the brightness of pedestrians is nonuniform. In the framework, an integrated decision can be made from the combination of the detection results in multiple frames. The detection rate of the system is greatly improved by the combination of SVM and the multiple object tracking with lower computation and is much higher than that of the normal systems.

Key words: pedestrian detection, multiple object tracking, Support Vector Machine (SVM)

中图分类号: 

• TP391