Vehicle Maneuver Recognition by Using Optical Flow

利用光流进行车辆机动识别

• 批准号: 12555151
• 负责人: SHIMIZU Eihan
• 金额: $ 4.48万
• 依托单位: The University of Tokyo
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (B)
• 财政年份: 2000
• 资助国家: 日本
• 起止时间: 2000 至 2002
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-12555151/
• 关键词: Optical Flow; Vehicle Maneuver; Moving Object Recognition; Sequential Image Analysis; Traffic Engineering; 実3次元空間計測; 交通事故分析; 挙動解析; 光学勾配法; 正則化

1. Review of optical flow estimation methodsOptical flow estimation methods in all pixels are divided into gradient-based approach and area-based approach. The basic methods of the gradient-based approach are increase in the number of observation equations and imposition of a condition. The area-based approach has cross correlation, sequential similarity detection, eigen window.2. Comparison of their performance empirically from the point of view of application to vehicle motion analysisThe reviewed methods were applied to sequential images from fixed video cameras and high altitude video cameras. The result showed the difficulty of estimation of precise and dense optical flow by ordinary gradient-based approaches. On the other hand, the results by SSDA were better than gradient-based approach.3. Development of vehicle recognition methodWith features, that are value of background subtraction and optical flow, all pixels in a spatio-temporal image are clustered. The spatio-temporal clustering means unifying pixels which meet homogeneous property.4. Application to real sequential imagesThe proposed method was applied to 10cm and 1/30sec sequential images. In this experiment 77 vehicles in all, whose movements include direct advances, left and right-hand turns, lane changes and stoppage, could be recognized.

1.光流估计方法综述全像素的光流估计方法分为基于梯度的方法和基于区域的方法。基于梯度的方法的基本方法是增加观测方程的数量和施加条件。基于区域的方法有互相关、顺序相似性检测、特征窗等。 2．从车辆运动分析应用的角度对它们的性能进行实证比较所审查的方法适用于来自固定摄像机和高空摄像机的连续图像。结果表明，普通的基于梯度的方法难以估计精确且密集的光流。另一方面，SSDA的结果优于基于梯度的方法。3．车辆识别方法的发展利用背景减法和光流值等特征，对时空图像中的所有像素进行聚类。时空聚类是指将满足同质性的像素点进行统一。 4．应用于真实连续图像所提出的方法适用于 10 厘米和 1/30 秒连续图像。本次实验共可识别77辆车，其运动包括直行、左转、右转、变道、停车等。

项目成果

布施孝志, 依田秀則, 清水英範: "高度撮影高分解能時系列画像を用いた車両挙動の認識に関する研究"日本写真測量学会平成13年度年次学術講演会発表論文集. 197-200 (2001)

Takashi Fuse、Hidenori Yoda、Hidenori Shimizu：“使用高分辨率时间序列图像识别车辆行为的研究”日本摄影测量学会 2001 年学术会议记录 197-200 (2001)。

Fuse, T., Yoda, H., Shimizu, E., Tsutsumi, M.: "A New Technique for Measurement of Traffic Flow on the Proposed High Altitude Platforms"Proceedings of 9th World Conference on Transportation Research. (CD-ROM). (2001)

Fuse, T.、Yoda, H.、Shimizu, E.、Ttsutsumi, M.：“拟议高空平台交通流量测量新技术”第九届世界交通研究会议论文集。

Fuse, T., Shimizu, E., Maeda, R.: "Development of Techniques for Vehicle Maneuvers Recognition with Sequential Images from High Altitude Platforms"ARIDA News Letter. 21. 1-2 (2002)

Fuse, T.、Shimizu, E.、Maeda, R.：“利用高空平台序列图像进行车辆机动识别技术的开发”ARIDA 新闻通讯。

Takashi FUSE, Hidenori YODA, Eihan SHIMIZU and Morito TSUTSUMI: "Auto-Recognition of Vehicle Maneuvers Based on Spatio-Temporal Clustering with Images from High Altitude Platforms"Proceedings of 3rd International Image Sensing Seminar on New Development i

Takashi FUSE、Hidenori YODA、Eihan SHIMIZU 和 Morito TSUTSUMI：“基于高空平台图像的时空聚类的车辆操纵自动识别”第三届国际图像传感新发展研讨会论文集

T.FUSE, H.YODA, E.SHIMIZU, M.TSUTSUMI: "Auto-Recognition of Vehicle Maneuvers Based on Spatio-Temporal Clustering with Images from High Altitude Platforms"Proceedings of 3rd International Image Sensing Seminar on New Development in Digital Photogrammetry.

T.FUSE、H.YODA、E.SHIMIZU、M.TSUTSUMI：“基于高空平台图像的时空聚类自动识别车辆机动”第三届数字摄影测量新发展国际图像传感研讨会论文集。