| 摘要: | 近年來，防盜的意識逐漸提高，各家廠商也不斷推陳出新，使得監控設備日益普及，在社區、大樓隨處可看見這些設備的身影。目前大樓的監視系統多半是透過多台攝影機安裝在各角落並使用不同角度來達到監控環境的目的，如此一來，不僅警衛要隨時注意每一個畫面的影像，且沒有統一空間的觀念，管理起來較不方便。
相較於一般視角的相機，魚眼相機的視角可高達 180 o，能拍攝到更寬廣的範圍；因此在相同的監控環境中，若使用魚眼相機可使設備數量大幅減少，進而降低系統建構及管理的成本。本研究使用魚眼相機作為主要的監控設備，提出整合式的全周監視與辨識系統；整個系統包含兩大部份：一是全周監視用於監視建築周遭的狀況，二是線上偵測前景物並分辨入侵者，若有異狀則透過物件辨識確認為入侵者後發出警示。
在全周監視部份，我們在建築的周圍各架設一台魚眼相機向下 25 o 拍攝，並在離線程序中利用所拍攝的多張不同入侵者位置，根據這些特徵點在影像上的座標求得地圖座標與影像之間的相對關係，再透過座標換算矩陣將物體的位置標示在地圖影像上，產生即時的全周監視影像。
為了避免不必要的警示，例如：車輛、貓狗經過…等，因此在本論文中加入了物件辨識系統，當偵測到前景物時，會擷取前景物的特徵做物件辨識確認為入侵者後，才會在畫面上顯示警示訊息通知使用者，藉此減少不必要的警示與人力資源的浪費。
本研究的系統使用兩部魚眼相機，並使用多段具有不同天候情況影片做實驗；在偵測方面，有 583 個樣本，平均有 96.5 % 的偵測率；在辨識方面，共有 681 個入侵者樣本，辨識正確率可達 93.8 %，誤判率為 2.53 %。錯誤辨識發生的原因是雨天的地面積水反射人影或前景物移動速度過快所造成的。
;In recent years, as the security awareness gradually increase and various manufacturers have been introducing new products, the monitoring devices become more prevalent. No matter where you are, you can see these devices, such as communities and buildings. Currently, most of the monitoring system of the building installation through multiple cameras at each corner and use different angles to achieve the purpose of monitoring the environment, as a result, not only to pay attention to guard image per screen at any time, and there is no concept of a unified space, and the management less convenient.
The view angle of a fisheye camera is 180 degree, so it can cover a wider field of view than a normal camera. Thus, in the same surveillance environment, only a few fisheye cameras can replace many traditional cameras to survey the events; such that the cost of system construction and management are then reduced. We use the fisheye camera as our main monitoring device, and propose integrated surrounding monitor and recognition system. The proposed system is composed of two major modules: surrounding monitor and online recognition system.
In the surrounding monitor module, we mount the cameras around the building and tilt 25 degrees. According to the relationship between the image plane and the surrounding map, we can solve the homography matrix and point the intruder on the surrounding map.
In recognition system, when a foreground object is detected, we extract the foreground object’s feature to recognize whether it’s intruder. If it were an intruder, the system will show alarm message on the screen to notice the user. Through the recognition system we can reduce most of unnecessary human resources.
We conducted experiments with the proposed system on several videos. The experiments results show that the average detection rate is 96.5 percent with 583 samples, the recognition rate can up to 93.8 percent with 681 samples and the average false positive rate is 2.53 percent. |
