Detection of Pedestrian Crossings with Hierarchical Learning Classifier from Multi-angle Low Altitude Images
-
摘要: 提出了一种基于特征的错误式学习分类器半自动迭代训练方法，该分类器能够自动识别多角度低空影像上的人行横道线，在人行横道线管控与数据库建立、道路网提取上有较好的应用。介绍了基于错误式学习的分类器训练思路与方法，并提出了将同一地区不同角度低空影像的识别结果进行合并，从而尽可能全面的检测出被城区高楼以及车辆遮挡的人行横道线的思路。通过对比实验该方法的鲁棒性，并在其基础上随机选取多组数据进行系列实验，证实了基于错误式学习的分类器比传统方式训练的分类器有更好的综合性能，能够在不降低识别耗时的前提下产生高准确率、低漏检率和低误识别率的识别结果。Abstract: This paper proposes a new training method for feature-based iterative hierarchical learning classifiers. It can be used to detect pedestrian crossings from multi-angle low altitude images. The training procedure and the method for merging multi-angle detection results are introduced in this paper. The performance of the classifier was evaluated based on random testing results. Experimental results from several datasets show that the iterative classifier has higher correctness, lower missing rate and lower error rate than the general classifier. Furthermore, the proposed method will not reduce the detection speed.
-
-
![]()
图 1 基于错误式学习的迭代训练流程图
Figure 1. Flowchart of Hierarchical Learning Based Training Process
![]()
图 2 遮挡、摄影角度对识别的影响
Figure 2. Interference to Detection Causing by Occlusions and Camera Angles
![]()
图 4 利用基于错误式学习的分类器对3张影像进行识别的结果
Figure 4. Detecting Results of the Hierarchical Learning Classifier
![]()
图 7 台湾省低空影像的人行横道识别结果
Figure 7. Hierarchical Learning Classifier Detecting Result of Taiwan Province
![]()
图 8 错误式学习与传统分类器性能对比图
Figure 8. Contrast in Quality of Detecting Between Hierarchical Learning Classifier and General Classifier
表 1 各组实验相机及影像参数
Table 1 Parameters of Cameras and Images
组别 拍摄地 相机型号 拍摄时间 焦距/mm ISO 影像质量 影像大小/像素 1 台湾 未知 2013-10 未知 未知 清晰 6 048×4 032 2 苏州 RICOH GXR 2013-03 18 200 轻微雾霾 4 288×2 848 3 未知 5D Mark 2 未知 24 400 清晰 5 616×3 744 
下载: 导出CSV
表 2 不同算法的识别结果各项指标对比
Table 2 The Contrast in the Quality of Detection of the Three Algorithm
算法 影像类别 M/% W/% A /% 训练耗时/h 识别耗时/min 神经网络模型 地面 2.652 14.952 93.746 - 358 圆形模板+SURF算法 低空 42.264 6.235 83.211 - 17 普通未迭代分类器 低空 15.746 69.778 85.450 约6 8 本文方法
(错误式学习分类器)低空 4.098 10.828 90.302 约28 8
下载: 导出CSV
-
[1] Sichelschmidt S, Sichelschmidt S, Haselhoff A, et al. Pedestrian Crossing Detecting as a Part of an Urban Pedestrian Safety System[C]. IEEE Intelligent Vehicles Symposium (Ⅳ), La Jolla, USA, 2010 http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5548032
[2] Boudet L, Midenet S. Pedestrian Crossing Detection Based on Evidential Fusion of Video-sensors[J]. Transportation Research Part C:Emerging Technologies, 2009, 17(5):484-497 doi: 10.1016/j.trc.2009.04.009
[3] Herumurti D, Uchimura K, Koutaki G, et al. Urban Road Network Extraction Based on Zebra Crossing Detection From a Very High Resolution RGB Aerial Image and DSM Data[C]. International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), Kyoto, Japan, 2013 http://dl.acm.org/citation.cfm?id=2573330
[4] Ishino Y, Saji H. Extraction of Road Markings from Aerial Images[C]. SICE Annual Conference, Tokyo, Japan, 2008 http://ieeexplore.ieee.org/document/4655024/
[5] Lienhart R, Maydt J. An Extended Set of Haar-Like Features for Rapid Object Detection[C]. International Conference on Image Processing, Rochester, USA, 2002 http://ieeexplore.ieee.org/xpls/icp.jsp?arnumber=1038171
[6] Schapire R E, Singer Y. Improved Boosting Algorithms Using Confidence-rated Predictions[J]. Machine Learning. 1999, 37(3):297-336 doi: 10.1023/A:1007614523901
[7] OpenCV Team. OpenCV 2. 4. 9 Documentation, 2014[OL]. http://docs.opencv.org/modules/objdetect/doc/cascade_classification.html, 2014
[8] 张祖勋.从数字摄影测量工作站(DPW)到数字摄影测量网格(DPGrid)[J].武汉大学学报·信息科学版, 2007, 32(07):565-571 http://ch.whu.edu.cn/CN/abstract/abstract1928.shtml Zhang Zuxun. From Digital Photogrammetry Workstation (DPW) to Digital Photogrammetry Grid (DPGrid)[J]. Geomatics and Information Science of Wuhan University, 2007, 32(07):565-571 http://ch.whu.edu.cn/CN/abstract/abstract1928.shtml
[9] Zhang Yongjun, Xiong Jinxin, Hao Lijuan. Photogrammetric Processing of Low-altitude Images Acquired by Unpiloted Aerial Vehicles[J]. The Photogrammetric Record, 2011, 26(134):190-211 doi: 10.1111/j.1477-9730.2011.00641.x
[10] 李德仁, 刘立坤, 邵振峰.集成倾斜航空摄影测量和地面移动测量技术的城市环境监测[J].武汉大学学报·信息科学版, 2015, 40(04):427-435 http://ch.whu.edu.cn/CN/abstract/abstract3223.shtml Li Deren, Liu Likun, Shao Zhenfeng. An Integration of Aerial Oblique Photogrammetry and Mobile Mapping System for Urban Geographical Conditions Monitoring[J]. Geomatics and Information Science of Wuhan University, 2015, 40(04):427-435 http://ch.whu.edu.cn/CN/abstract/abstract3223.shtml
计量
- 文章访问数: 1335
- HTML全文浏览量: 129
- PDF下载量: 270
