在視訊監控和機器人視覺應用，物件追蹤扮演了重要的角色。典型的物件追蹤演算法需要大量計算資源，導致在資源受限的嵌入式系統實現即時物件追蹤的困難。本論文致力於設計一個平行化架構的物件追蹤硬體加速器。此系統包含了特徵擷取模組、預估位置模組及PSO追蹤模組，以及最上層的管線控制器。特徵擷取模組利用灰度統計和哈爾特徵建立多特徵聯合稀疏矩陣，作為追蹤物的樣板資訊，接著預估搜尋範圍，最後粒子群最佳化追蹤物件移動。我們採用開放資料庫進行驗證和測試。實驗結果顯示，我們的系統能夠滿足即時追蹤的性能需求。同時與先前的研究結果比較，本系統減少了34%的硬體資源使用。;Object tracking has been a popular application in computer vision, for example,public area surveillance, and robot vision, etc. Due to typical object tracking algorithm needs high-efficiency hardware resources to reduce the processing time, it is difficult to implement a real-time object tracking in resource-constrained embedded systems. In this paper, we design a parallel architecture object tracking hardware accelerators. The architecture of the accelerator contains Feature module, Prediction module, PSO tracking module and a top layer pipeline controller. Feature module constructs multi-feature joint sparse matrix by using grayscale statistics and Haar-like features, and uses them as the template of the tracking object. Then, estimates the searching scale. Finally, Particle Swarm Optimization (PSO) is used for tracking object’s movement. We adopt open source database to verify and test our modules. Experimental result shows that our system can satisfy real-time object tracking requirement. Comparing with previous research consequence, our system reduces 34% usage of hardware resource.
