為了提升追蹤目標物的速度和解決目標物被遮蔽的問題，本論文採用改良的粒子最佳化演算法“自組織隨時調變係數的粒子最佳化演算法”來對目標物進行追蹤。 HPSO-TVAC演算法主要利用族群之間各個成員的相互關係，使族群整體朝向更好的目標前進的一種演算法，並利用改良的自適應搜尋框，使搜尋框大小在追蹤目標追蹤不到的時候變大，在一直能追蹤到追蹤目標時，保持較小的大小，使粒子們能搜尋得更加精確，並能有效解決遇到遮蔽物的情況。 改良的種子區域成長法，主要是改良產生種子的部分，使原本標記出來需要重新搜尋四鄰位置的種子數量減少，其目的是讓我們區分不同的目標物，並使各個目標物各自連成一塊，使我們可以算出各個目標的中心位置。 本論文使用背景相減法以分離背景和移動目標，改良的種子區域生長法區分各個不連通的目標物，並且算出各個目標的中心位置，利用顏色直方圖來建構目標物的模型，以HPSO-TVAC來對各個目標物進行追蹤。最後比較其他不同的演算法進行模擬測試，並且將HPSO-TVAC實際測試於即時的多目標追蹤。 ;In order to improve the tracking speed and solve the shadowing problem, this paper trace objects by “Self-Organizing Hierarchical Particle Swarm Optimizer with Time-Varying Acceleration Coefficients (HPSO-TVAC)”, which improved from PSO algorithm. We use the relationship between the members of the groups to make whole groups go along toward the way of better goals by HPSO-TVAC algorithm, and we also uses “adaptive searching window”. Then the searching window will zoom in or out which depends on global best fitness. While we can find the targets, we will make the searching window contains small, but while we cannot find the targets, we will let the searching window bigger to find the targets. The improved seeded region growing method, which we mainly improved the quantity of seeds, is presented in this study. We reduce quantity of seeds to increase the efficiency, and it can let us distinguish between different targets. In this study, we use background subtraction to distinguish background and moving objects, and we also use improved seeded region growing method to distinguish different targets. Then we use color histograms to build target models, and trace every targets by HPSO-TVAC algorithm.
