寬闊環境下的行人監控系統之研究

Abstract

將攝影機擺放在距離行人較遠處，可以監控到更廣泛的環境，並減少攝影機的數量，達到成本降低的目的，但是在這種寬闊環境或大範圍監控下，行人面積相對整體影像面積甚小，意味著能夠從行人上獲取的資訊量減少，且追蹤時受到背景的干擾程度也會提升，導致追蹤結果不理想甚至失敗。本論文提出一種區塊色彩表面模型，並且結合粒子濾波器等技術，能有效的屏除背景像素在追蹤時的干擾，得到更準確的結果。 本論文貢獻有以下兩點，第一點，本論文改良區塊表面模型，提出區塊色彩表面模型，能夠消除背景的干擾、修飾不完整的前景、處理遮蔽情形與二次分類行人，使得追蹤準確性提高；第二點，本論文利用區塊色彩表面模型結合粒子濾波器、方向梯度直方圖與資料關聯等技術，建構出寬闊環境下的行人監控系統之雛型。 關鍵字：寬闊環境、多行人追蹤、粒子濾波器與目標物模型

Setting camera far away in the intelligent surveillance system can not only surveil wider environment, but also reduce the required numbers of cameras, achieving the goal of costing down. At the environment like this, so called wide environment, the size and area of pedestrians, comparing to overall image, are much smaller. It means the quantity of information extracted from pedestrians decrease, and the effect of background increase, resulting in bad tracking result. In order to avoid the effect of background and make tracking result more accurate, this paper propose patch color base appearance model. There are two contributions as follow. First, this paper proposes a more robust appearance model, patch color based appearance model; second, this paper proposes a prototype of multiple pedestrian surveillance system based on patch-color-based appearance model with particle filter, histogram of gradient orientation and data association. Keywords: Wide environment, pedestrians tracking, particle filter and target model