監控系統廣泛的應用在日常生活中，不過目前所使用的監控系統 大都只有錄影功能，無法立即針對異常事件發出警訊，所以有人提出 智慧型監控系統的概念，利用電腦視覺的方式，在不需要人為操作的 方式下，讓監控系統能自動分析攝影機擷取的影像，使其具有偵測、 追蹤與辨識等功能。 本論文提出將前景偵測實現在嵌入式雙核心平台上，可以即時且 有效的偵測出前景物。前景偵測演算法的部份，我們是採用計算量低，記憶體使用量少的CodeBook 演算法；嵌入式平台實現的部分包含：軟體上利用Codec Engine 架構的方法，搭配在硬體上利用共享記憶體(share memory)的配置，來實現ARM 與DSP 彼此間資料的傳遞；在定點式DSP 上將程式定點化，來提高DSP 在處理資料上的效率；利用嵌入式平台雙核心特性設計多執行緒，來達到平行處理的效果。實驗結果可以顯示，本論文可以即時且有效的偵測出前景物，在 執行速度上每秒可以處理28 至30 張幀(frame)。Surveillance systems are widely applied in our daily life. However, the main function of current surveillance systems only focuses on video data recording. Besides, human has to pay tremendous attention in monitoring the video data and issuing alarm when abnormal events happen. To save the human resource, an automatic and intelligent surveillance system is developed in this thesis to detect, track, recognize and analyze moving objects. The main goal of this thesis is to implement the task of foreground detection on a dual-core embedded platform. A codebook-based algorithm is developed to perform the moving object detection task in resource lacking workspace. Hence, lower computational power and less memory space are needed in performing the detection task. The main task in the dual-core system is to establish a reliable communication between ARM and DSP. A software-based core engine framework is adopted to allocate the share memory and support the hardware. Due to the non-supporting of floating-point arithmetic hardware in DSP, the fixed-point operation is performed to make DSP work efficiently. The resources are well allocated to execute the parallel process on the dual-core platform. The system is installed in two places to evaluate the performance. Experimental results demonstrate that the implemented system can accurately and efficiently detect the moving objects in real time.
