- 學位論文
結合Cam Shift和Kalman filter運用在物件追蹤
Dynamic Object Tracking based on CamShift and Kalman Filter
摘要
本篇主要探討物件在追蹤導向過程中,藉由結合Cam Shift理論和Kalman Filter來達到系統整體追蹤上的穩定性。在一般情況下運用Cam Shift理論的追蹤系統,僅限於在單一顏色背景且低色相的物件,本文提出運用Kalman Filter理論上針對動態物件追蹤上的改善與採樣時間縮短等優處增加系統的穩定度。 探討過程中,首先將我們的目標圖件從RGB圖案進行HSV圖像空間轉換成直方圖和倒向投影部分進行初始設置,再來從上面得到的初始數值圖進行搜索窗口大小計算以及窗口內的質心位置,實驗過程中藉由系統運行Cam Shift理論,使我們的窗口中心持續移動至質心直到移動距離小於我們的固定閥值距離或是運算次數達到最大值止,藉由數次得到的結果延續使用到下一圖像序列中得到我們連續追蹤過程。實驗過程後我們可以得知,結合Cam Shift理論和Kalman Filter的物件追蹤系統比起單一理論來的穩固,也能突破單一理論上受限環境上的條件。 此次研究貢獻主要有以下三點: (1) Kalman濾波器改善了我們系統在複雜環境下,雜訊和動態物件追蹤不丟失性。 (2) Cam Shift雖然能改善Mean shift當目標尺寸變化導致尺度準位不準確狀況,針對複雜環境下容易受到干擾,實驗後可發現Kalman Filter可以明顯改善結果。 (3) 我們提出的改良系統能讓單一追蹤系統測量範圍提升到動態物件上的追蹤部分加以改善
並列摘要
In the thesis project of object tracking, we propose the thesis combining the Cam Shift algorithm with Kalman filter which can provide the position of the target more accurately. The tracking algorithm of Cam Shift can only be used in simple background with few colors. By combining the Kalman filter, we can reduce the interference of the background color and make the tracking more accurate. In this thesis, we follow these steps: First, by converting our target image into the H.S.V. color space, we can get desired color information and adjust the contrast of the hue, saturation and brightness. Then we need an initial point to start our tracking process. The traditional Mean Shift theorem does not allow users to change its window size and may lose its target, so that we calculate the search window size and the position of the mass center based on the Cam Shift algorithm. The process is continued until the desired accuracy is met. Finally, due to the instability of the complex background, we add the Kalman filter to lower the noise and interference during the experiment. After the process, we can achieve higher accuracy. The experiments show that the Cam Shift algorithm with the Kalman filter can reduce the noise and instability much more effectively than using only the former. In this thesis, the contributions of the research are as follows: 1.Kalman filter improves and strengthens the stability of the Cam Shift algorithm execution in complex background. 2.It is difficult to detect the object with various colors using in Cam Shift algorithm. Combining it with the Kalman filter improves the object detection capability and make the result more precise. 3.The object tracking experiment helped us expand the measurement range from single color still object to dynamic background processing.