行人检测在汽车驾驶辅助系统和智能视频监控等领域有广泛的应用，而行人候选框的生成是行人识别、定位及跟踪的一项重要前期工作.本文提出一种基于区域复合概率（Local Mixture Probability，LMP）模型的在线生成行人候选框的方法.该方法根据区域相似性将监控场景划分为多个子区域，随之对各区域内行人的位置和尺度分别建立泊松模型和高斯模型.通过模型的学习与更新可以获取目标出现的概率信息以及目标尺度的分布情况，从而为候选框的生成提供依据，避免遍历搜索的盲目性.实验结果表明，该算法能够在生成较少数目候选框的情况下获得较高的覆盖率.
Abstract
Pedestrian detection is widely applied in driver assistance systems and video surveillance fields,while proposal generation is a significant preliminary work for pedestrian recognition and tracking.This paper proposes a method for fast online proposal generation using Local Mixture Probability (LMP) model.Poisson model and Gaussian model are separately established for online learning location and scale of pedestrians after region-dependent segmentation according to local similarity.Based on learning and updating models,both the probability of pedestrians occurrence and the probability distribution of the scale in specific regions can be obtained,which provides bases for pedestrian proposal generation and avoids searching blindness.Experiments on Caltech Pedestrian dataset show that LMP can achieve higher recall by fewer pedestrian detection proposals.
关键词
机器视觉 /
行人检测 /
区域复合概率模型 /
生成候选框
{{custom_keyword}} /
Key words
machine vision /
pedestrian detection /
LMP model /
proposal generation
{{custom_keyword}} /
中图分类号：
TP391.4
{{custom_clc.code}}
({{custom_clc.text}})
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 杨涛,李静,等.基于场景模型与统计学习的鲁棒行人检测算法[J].自动化学报,2010,36(4):499-508. YANG Tao,LI Jing,et al.Scene modeling and statistical learning based robust pedestrian detection algorithm[J].Acta Automatica Sinica,2010,36(4):499-508.(in Chinese)
[2] HOSANG J,BENENSON R,et al.What makes for effective detection proposals?[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2015,38(4):814-830.
[3] CHENG M M,ZHANG Z,et al.BING:Binarized normed gradients for objectness estimation at 300fps[A].IEEE Conference on Computer Vision and Pattern Recognition[C].US:IEEE,2014.3286-3293.
[4] ZITNICK C L,DOLLAR P.Edge boxes:Locating object proposals from edges[A].13th European Conference on Computer Vision[C].Zurich,Switzerland,2014.391-405.
[5] HORN B K P,SCHUNCK B G.Determining optical flow[J].Artificial Intelligence,1981,17(1-3):185-203.
[6] TERZOPOULOS D.Regularization of inverse visual problems involving discontinulties[J].IEEE Trans PAMI,1986,8(4):413-424.
[7] LUCAS B,KANADE T.An iterative image registration technique with an application to stereo vision[A].International Joint Conference on Artificial Intelligence[C].Canada,1981.674-679.
[8] GERONIMO D,LOPEZ A M.Vision-Based Pedestrian Protection Systems for Intelligent Vehicles[M].NewYork:Springer Publishing Company,2013.
[9] UIJLINGS J R,SANDE K E,et al.selective search for object recognition[J].International Journal of Computer Vision,2013,104 (2):154-171.
[10] GIRSHICK R,DONAHUE J,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[A].27th IEEE Conference on Computer Vision and Pattern Recognition[C].US:IEEE,2014.580-587.
[11] GIRSHICK R.Fast R-CNN[A].15th IEEE International Conference on Computer Vision[C],Santiago,Chile,2016.1440-1448.
[12] REN S,HE K,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[A].29th Annual Conference on Neural Information Processing Systems[C].Montreal,QC,Canada,2015.91-99.
[13] 覃剑,王美华.采用在线高斯模型的行人检测候选框快速生成方法[J].光学学报,2016,36(11):162-172. QIN Jian,WANG Mei-Hua.Fast pedestrian proposal generation algorithm using online Gaussian model[J].Acta OpticaSinica,2016,36(11):162-172.(in Chinese)
[14] DOLLAR P,WOJEK C,et al.Pedestrian detection:an evaluation of the state of the art[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2012,34(4):743-761.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
教育部-中国移动科研基金项目 (No.MCM20150102）; 重庆高校创新团队建设计划 (No.CXTDX201601006）
{{custom_fund}}
PDF(1341 KB)
