近年來，隨著攝影機與監視器的普及，物件偵測及物件追蹤是電腦視覺領域中一門重要且充滿挑戰性的研究課題。針對單一物件追蹤而言，其困難點在於複雜環境與追蹤物件的高變異性，複雜環境因素包含：光線、角度，而追蹤物外觀的變異性又可分為剛體（角度變化）與非剛體形變，以及遮蔽等問題。 本研究中，主要應用改良型PSO演算法(PMHPSO-TVAC)來對目標物進行即時追蹤。而在偵測目標物方面，則以影像相減法來切割出目標物與背景。再來則是利用改良的種子區域生長法來標記各個目標物，區分出各個目標物後，再計算出各個目標物的中心位置。接著對各個目標物建構顏色直方圖與深度直方圖模型以便做追蹤使用，然而在追蹤過程中很容易受到光線變化影響，採用HSV色彩空間中的色相，盡量減少了亮度的影響。然而，在光線極度昏暗的情況下能無法改善干擾，故本論文建構目標物的深度直方圖模型來補償目標物的描述方式。 最後，利用目標物的深度直方圖與顏色直方圖模型，以PMHPSO-TVAC演算法來進行多目標追蹤。 ;In recent years, with the popularity of the camera and monitor, object detection and object tracking field are important and challenging research topic. For object tracking, it is difficult to track objects in complex environments. In order to improve the tracking speed and solve the shadowing problem, this paper uses Position Mutated Hierarchical Particle Swarm Optimization with Time-Varying Acceleration Coefficients (PMHPSO-TVAC) algorithm for object tracking in real time. In terms of object detection, in this study, the background subtraction is used. And can cut out complete targets. The background subtraction has low computation and be easily applied to real-time systems. Besides, the improved seed region growing method is used to distinguish every target. Then, for model building, color histograms are used to build target models. However, in no-light environment, we can’t track any target, so this paper construct depth histogram object model to compensate for object model. Finally, we used the depth histogram and color histogram model with PMHPSO-TVAC algorithms for multi-target tracking.
