摘要
针对行人检测存在识别精度不高,实时性较差等问题进行相关研究。分析了基于多尺度滑动窗口法提取行人检测窗口的缺点,为解决行人检测中检测窗口数量过多的问题,提出在图像分割和路面提取的基础上实现对行人检测窗口的提取。先利用FCM聚类算法训练得到分割阈值,其次提取路面区域,根据路面区域筛选可能存在的行人位置,进而提取感兴趣区域,并对相应的感兴趣区域提取HOG特征进行进一步精确分类。实验结果表明,采用基于路面约束的图像分割方法来提取感兴趣区域,有效减少了遍历窗口的数量,从而提高了行人检测速度和检测精度。
The article discuses the existing problems of pedestrian detection, including low recognition rate, poor real-time per- formance. The shortages in sliding window algorithm are analyzed. In order to reduce the number of the detection window to im- prove the speed of recognition, this paper proposes a method of image segmentation based on the road constraint and extracts the re- gion of interest （ROI）. Firstly segmentation thresholds are obtained by FCM clustering algorithm, and road surface is extracted by region growing algorithm. According to the road area estimate the possible pedestrian＇ s position and then extract the interested re- gion. Finally further accurate detection of pedestrian is implemented by extracting HOG feature. Comparing with the sliding window method, the experimental results show that extraction of ROI based on road surface constraint and the image segmentation method can effectively reduces the number of traversal of the window and improves the detection speed and detection accuracy.
出处
《电视技术》
北大核心
2017年第4期239-242,共4页
Video Engineering
基金
吉林省科技发展计划项目(20140204018GX)
关键词
行人检测
感兴趣区提取
图像分割
路面约束
HOG
Pedestrian detection
Regions of interest （ROI）
Image segmentation
Road surface constraint, HOG
