強健而即時的自動追蹤燈系統之研發

本研究發展一套舞台自動追蹤燈系統，提出強健而即時的追蹤演算法，並改善目前使用超音波技術昂貴與應用上諸多限制等缺點，具有經濟、節省人力、易於架設和操作之優點。系統分為影像擷取、追蹤與追蹤燈三部份。首先以機器視覺系統擷取彩色影像，操作者以滑鼠框選目標演員上下半身，接著以個人電腦進行影像處理與分析，得到演員運動資訊。追蹤階段則結合二種演算法，為本研究最重要的部份：上下半身各設計一組互鎖雙特徵均值移動演算法，與傳統均值移動(mean shift)演算法相較，多考慮了上下半身連鎖距離之特徵；另一重點為動態更新卡爾曼濾波器，將均值移動演算法計算得知的位置、速度與加速度狀態向量送入卡爾曼濾波器進行預測，考慮系統模型誤差與量測模型誤差，即時更新系統雜訊共變異矩陣與量測雜訊共變異矩陣，達到更佳的預測效果，同時以當時的預測位置作為均值移動演算法新的搜尋起點，並根據預測速度調整搜尋視窗大小。最後將預測資訊送至追蹤燈部份，經由步進馬達轉動以控制燈光照射在演員身上。實驗結果證實，若使用傳統單特徵均值移動演算法，當二位演員之某半身顏色相同且交錯時往往造成追蹤失敗，但以互鎖雙特徵均值移動演算法則可克服追蹤失敗的情形。追蹤之速度可達111.3像素/秒，可因應各種舞台表演之移動速度，符合即時運算之優點；即時更新卡爾曼濾波器動態地更新系統雜訊共變異矩陣與量測雜訊共變異矩陣，可大幅提升追蹤成功率，使卡爾曼濾波器的預測更為強健。本研究所提出之影像追蹤演算法具有強健而即時之優點，能識別、預測與追蹤舞台演員，可應用於真實舞台上的表演。

關鍵字

即時影像追蹤；機器視覺；均值移動演算法；卡爾曼濾波器；模型；估測；追蹤燈；自動化

並列摘要

An automated followspot tracking system with a robust and real-time tracking algorithm was developed. The proposed image tracking algorithm has two advantages: robustness and real-time processing. Our approach is divided into three parts: image acquisition, tracking, and spot following. In the beginning, image sequences are captured by a machine vision system. Operators select the upper body and the lower body of an actor to be tracked. The actor’s movement information is extracted through a series of image processings and then sent to the tracking part. There are two most important methods in our tracking algorithm. The first method is the interlocked dual-feature mean shift which considers both color and spatial difference features. We adopt this method because failure occurs when we use the mean shift approach with only one single feature, which considers only color feature, if the actor passes by another actor in the same color of his half body. However, the interlocked dual-feature mean shift could overcome the shortcoming. The second method is the updated Kalman filters which estimate system noise and measurement noise, and then update the corresponding parameters. The estimated states consist of position, velocity and acceleration of the actor. The estimated position is sent not only to the followspot controller for lighting but also back to the interlocked dual-feature mean shift for updating the centers of searching regions in the next sampling time. Also, the estimated velocity is employed to adjust dynamically the searching region. Results show that the approach with the updated Kalman filter is more robust than the non-updated one because the rate of successful tracking is highly promoted. The maximum tracking velocity is up to 111.3 pixel/sec, which is sufficient for tracking characters on stage. The algorithm improves the disadvantages of current followspot system, such as high price and inconvenience. Consequently, the developed system is more economical, manpower-saving, and easy to setup and operate.

並列關鍵字

real-time video tracking ； machine vision ； mean shift ； Kalman filter ； model ； estimation ； followspot ； automation

參考文獻

2. 林振翔。2005。舞台人物自動追蹤燈系統之研發。碩士論文。台北：台灣大學生物產業機電工程學研究所。

5. AU1210 規格表. Available at:

6. Baumberg, A. and D. C Hogg. 1994. An efficient method for contour tracking using active shape models. In “Proc. IEEE Workshop on Motion of Non-rigid and Articulated Objects”, 194-199.

8. Collins, R. T., A. J. Lipton, H. Fujiyoshi, and T. Kanade. 2001. Algorithms for cooperative multisensor surveillance. In “Proc IEEE”, 89(10):1456-1477.

9. Comaniciu, D., V. Ramesh, and P. Meer. 2000. Real-time tracking of non-rigid objects using mean shift. In “Proc. IEEE Conference on Vision and Pattern Recognition”, 2(2):142-149.

國際替代計量

強健而即時的自動追蹤燈系統之研發

全文下載

主題瀏覽