基於鄰近物體互動力模型之擁擠交通路口環境多目標追蹤系統

Multiple Target Tracking Using Neighboring Interaction Force Models at Crowded Traffic Intersections

林國輝(Kuo-Huei Lin)

指導教授 ： 王傑智

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2011年)

https://doi.org/10.6342/NTU.2011.00987

摘要

可靠的交通路口監測方法，能夠在許多智慧型運輸系統(intelligent transportation system, ITS)的任務中，提供關鍵的資訊。考量安全因素，各式各樣的物體，例如行人、自行車、機車、小客車、和巴士，必須被可靠的追蹤。然而在擁擠的交通路口，多目標追蹤(multiple target tracking)是一項艱鉅的任務。高度的擁擠導致具挑戰性的資料連結(data association)問題。
本論文提出一種新方法，用高層的運動模型來解決多目標追蹤具挑戰性的資料連結問題。鄰近物體互動力模型(the neighboring interaction force models, NIF)被提出用來考慮被追蹤目標和他的鄰近物體之間的短期互動。藉由動態結構多模型(variable structure multiple model, VSMM)的架構，整合了這些互動力模型、等速度運動模型(the constant velocity model, CV)和等加速度運動模型(the constant acceleration model, CA)。
本文提出的方法，以擁擠交通路口所取得的平面光雷達測量資料來驗證。這些資料經由人工標記，用來評估所提出方法與系統的效能。實驗結果顯示，在擁擠環境中，目標被遮蔽或是運動變化劇烈的困難情況，本文提出的方法，相較於線性運動模型(the linear maneuver motion model)、場景互動模型(the scene interaction model)、和鄰近物體互動模型(the neighboring object interaction model)，分別改進了27.0%、14.8%、和18.9%的效能。

關鍵字

智慧型運輸系統 ； 光雷達追蹤 ； 多目標追蹤 ； 動態結構多模型 ； 鄰近物體互動力模型 ； 資料連結

並列摘要

Robust traffic intersection monitoring provides critical information to a number of intelligent transportation system-related tasks. A wide variety of moving objects such as pedestrians, bicycles, motorcycles, cars and buses must be tracked robustly for safety. However, multiple target tracking at crowded traffic intersections is a daunting task. High crowdedness causes challenging data association problems. This work presents a new solution using higher level motion models to solve challenging data association problems in multiple target tracking. The neighboring interaction force models are proposed to take short-term interactions between a tracked object and its neighboring objects into account. The interaction force models are seamlessly integrated with the constant velocity model, and the constant acceleration model under the existing variable-structure multiple-model (VSMM) framework. The proposed approach was validated using 2D LADAR measurements collected at crowded traffic intersections. The human annotated data were applied for evaluating the proposed approach and system. The ample experimental results show that the proposed framework outperforms the linear maneuver motion model, the scene interacting model, and the neighboring object interaction model approaches for 27.0%, 14.8%, and 18.9% respectively in difficult situations with occlusion or rapid change motion in crowded scenes.

並列關鍵字

Intelligent Transportation System ； LADAR-based Tracking ； Multiple Target Tracking ； Variable Structure Multiple Model ； Neighboring Interaction Force Model ； Data Association

參考文獻

Beymer, D., McLauchlan, P., Coifman, B., & Malik, J. (1997). A real-time computer vision system for measuring traffic parameters. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition, San Juan, Puerto Rico.

Blom, H. A. P. & Bar-Shalom, Y. (1988). The interacting multiple model algorithm for systems with markovian switching coefficients. IEEE Trans. On Automatic Control, 33(8), 780–783.

Chatterjee, S. & Hadi, A. S. (1986). Influential observations, high leverage points, and outliers in linear regression. Statist. Sci., 1, 379–416.

Cox, I. J. & Hingorani, S. L. (1996). An efficient implemenation of Reid’s multiple hypothesis tracking algorithm and its evaluation for the purpose of visual tracking. IEEE Transaction on Pattern Analysis and Machine intelligence, 18(2), 138–150.

Cui, J., Zha, H., Zhao, H., & Shibasaki, R. (2007). Laser-based detection and tracking of multiple people in crowds. Computer Vision and Image Understanding, 106(2-3), 300–312.