基于似物目标的快速行人检测算法-ChinaXiv.PDFVIP

/article/02-2019-08-054.html 计算机应用研究 基于似物目标的快速行人检测算法 * 1, 2 1 1 刘 倩 ，李 策 ，杨 峰 (1. 中国矿业大学(北京) 机电与信息工程学院, 北京 100083; 2. 宁夏大学 信息工程学院, 银川 750001) 摘 要：针对传统滑动窗口行人检测速度慢、无效窗口数量大和识别率低的问题，提出基于似物目标的快速行人检测 算法。该算法首先使用改进的二值化赋范梯度算法生成目标候选集，再利用提出的NDOG 特征描述子计算目标特征。 通过在INRIA 和Caltech-USA 两个行人数据库分析实验，证明该算法可有效减少待检测窗口数量，提高特征表示能力， 且计算时间显著降低，使用线性支持向量机作为分类器可获得较高的行人检测率。 关键词：似物目标检测；二值化赋范梯度；方向梯度直方图；极限学习机 中图分类号：TP391.4 doi: 10.3969/j.issn.1001-3695.2018.01.0115 Rapid pedestrian detection based on generic object generation 1 1, 2 1 1 v Liu Qian , Li Ce , Yang Feng 9 (1. School of Mechanical Electronic Information Engineering , China University of Mining Technology, Beijing 10083, 6 3 China; 2. College of Information Engineering Ningxia University, YinChuan 750001, China) 2 0 Abstract: Detection of pedestrians on the image using a sliding window mode results in a huge amount of invalid windows and . 4 inefficiency. To circumvent these problems for pedestrian detection, this paper proposed an approach, based on generic object 0 generation, for rapid detection of pedestrian. Firstly, the approach generated the generic object proposals by the Improved 8 1 Binarized Normed Gradients, and then calculated the character of proposals based on our NDOG feature descriptor. 0 Experimental results over the INRIA and Caltech-USA pedestrian datasets show that our approach reduces the number of 2 : proposals effectively, and improves the efficiency and the detection accuracy signif