使用影像處理技術於即時人物移動追蹤與路徑預測之研究

林衍百 ； 柯秉廷 ； 吳俊霖 ； 吳逢騏

《TANET2017 臺灣網際網路研討會》 2017年 (2017/10) Pp. 364-369

https://doi.org/10.6728/TANET.201710.0065

摘要

本論文成功實現在固定攝影監控下能夠持續追蹤移動的物體並持續更新物體的資訊，接著算出物體的移動軌跡並計算物體的速度，最後透過歷史資訊來預測物體路徑並預測兩物體是否即將碰撞。為擷取物體所在的位置，本論文利用背景相減的辦法，計算出每個影像與背景的差異區塊，並持續更新背景。由於要取得完整的物體，本論文利用膨脹侵蝕等方法來消除雜訊，並使用繪製輪廓、凸殼(convexhull)等影像處理方法來填補物體中的空隙，最後使物體成為一個單純的滴狀物(blob)。為了達成即時處理影像並簡化運算，本論文放棄對物體做特徵標記。利用物體過去的資訊做加權平均求得預測位置，並比對物體的實際與預測位置之距離來配對，以此進行同一物體之追蹤。最後以物體重心繪製軌跡線並求得瞬時速度，若在一段時間內物體速度超過一定值，則給予超速警告，若單一物體超速且兩物體之預測位置過近，則給予碰撞警告。實驗結果顯示所提演算法在較單純影像背景皆能正確執行，持續標記追蹤目標且更新資訊，且不受限於室內與戶外背景的影響。

關鍵字

即時監視系統 ； 預測路徑 ； 碰撞警告 ； 影像處理 ； 形態學

並列摘要

This paper successfully achieved in photography under the fixed monitor to keep track of moving objects and continuously updated the information of objects, then calculate the movement and speed of objects. Finally, we predict the path of objects and predict whether two objects are about to collide through the historical information of objects. In order to acquire the location of the objects, we used the method of background subtraction to calculate the difference of each image and the background blocks of each image, and continuously updated the background of each image. Because we want to get the whole object, we used dilation, erosion and other methods to eliminate the noise of the images were calculated by the method of background subtraction, and used to draw the outline, convexhull and other image processing method to fill the interstices in the object, and at last we make the object into a pure blob. To achieve real-time processing images and simplify operations, this paper gave up on tagging feature on the objects. We used the past information of the objects to calculate the weighted average and obtain the predicted place, and compare the distance of the actual and predicted place to matching. Finally we drew the trajectory line with the center of gravity to obtain and the instantaneous velocity Center of gravity of the object, if the speed exceeds a certain value of an object over time to give the speed warning, if a single object speed and too close to predict location of two objects, the collision warning. Experimental results show that the proposed algorithm in a simpler background image can execute correctly, continuously marked target and updated information, and is not limited to effects of indoor and outdoor setting.

並列關鍵字

real-time monitoring system ； the predicted path ； collision warning ； image processing ； morphology