Algorithms and Chip Design for Video Moving Object Segmentation in Fixed Video Camera

[[abstract]]隨著新的影像壓縮技術的發展，物件導向設計的壓縮模式〈object-based coding〉漸受重視。但必須事先自影像序列當中取出各個物件，而動態影像切割技術便是取得影像序列當中物體的方法。若是動態影像切割技術能提供準確的影像切割結果，物件導向設計的壓縮模式的效率便可以有效的提升。 所以在本論文中將針對固定式攝影頭所取進來的視訊序列，做視訊物件的切割。切割方法可分為兩大部分：一為動態物件動態特性找出粗略邊緣、再用空間信號處理物件邊框做修整。首先我們利用調適性方法在時間域上找出相關性足夠的參考圖框，再以調適性的量化及濾波方法，找出物件的粗略外廓。接著我們將已找出的外廓做低通濾波處理與量化使外廓更平順，再加以修正物件的外廓取得較精細的物件外廓。 根據此演算法，我們設計成硬體架構。使用硬體描述語言(Verilog)，配合Cell_Based流程加以實現。晶片面積大小 mm2，內含兩條352x8 Line Buffer與一條352x2 Line Buffer，最高工作頻率為50MHz，預估消耗功率175.60mW。[[abstract]]The content-based video coding, MPEG-4, is a new standard for the next generation system. Video segmentation is a very important unit. In this thesis, a new moving object segmentation algorithm and its chip design are proposed. First, we adaptively compute the temporal differences of the current and reference frames using bi-direction method to find the basic moving object information. Second, an adaptive truncation processing is used to remove background content as well as to discard background noise to achieve a binary shape. Third, a boundary smoothing method is presented to improve the segmentation quality. Based on the moving object segmentation algorithm, a real-time VLSI architecture is developed using cell-based design