| 摘要: | 監控設備在日常生活中隨處可見，無論是在公共場所或私人空間，如學校、公司、賣場、街道路口、家庭等，都是視訊監控的應用範圍。然而，一台相機可拍攝的範圍有限且容易受到環境結構遮蔽的影響，難以涵蓋所要監視的範圍，因此多相機聯合偵測系統的發展越來越受關注。
相較於一般視角的相機，魚眼相機的視角可高達 180 o，可拍攝到更寬廣的範圍；因此在相同的監控環境中，使用魚眼相機可減少設備的使用量，降低系統建構及管理的成本。本研究使用兩台魚眼相機為例，提出一個自動偵測及追蹤的環境監視系統 (surveillance system)；整個系統共分為兩部份:前景物偵測與前景物追蹤。
在前景物偵測上，我們將相機固定在室內天花板下，以背景相減法 (background subtraction) 偵測前景物。以一段時間訓練背景，利用新進影像和背景影像的差異，檢測出畫面的前景物。根據背景亮度較低但色彩差異不大的特性處理陰影問題，減少錯誤前景物的判斷；使用形態學 ( morphology) 處理畫面雜訊點，對於急遽的光線變化如開關燈和緩和的光線變化如日光漸變可以自動更新背景。
在前景物追蹤上，我們使用卡爾曼濾波器 (Kalman filter) 預測前景物的位置，在相機重疊畫面預先建立轉換表，當前景物跨越相機監視範圍時，利用轉換表能了解不同相機追蹤的物件其實是同一前景物，維持物件編號聯合追蹤。為了要增加比對的可靠性，加入前景物的外型特徵輔助判斷；例如，顏色。在系統運作時兩部相機皆可實現多目標的追蹤，並且持有各自的追蹤器，當物件發生短暫遮蔽仍持續預測位置並提供平順的移動軌跡。
本研究的系統使用兩部魚眼相機，使用多段具有不同光線亮度和人物數量的影片做實驗，平均有96.7 %的敏感度，誤判率為0.45%，錯誤判斷發生的原因是前景物的衣物和背景色彩很相似，加入Kalman filter追蹤輔助後可以使敏感度提升為98.55 %。實驗結果證實現了我們的系統能適應室內監控許多挑戰狀況，如光線變化、陰影干擾、遮蔽，是有效且穩定的聯合前景偵測和追蹤系統。
;Video surveillance has widely applied in our daily life, both in public and private environments, such as schools, offices, shopping malls, streets, and homes; however,the view scope of a single camera is finite and limited by scene structures. In order to monitor a wide area and trace a complete trajectory of a moving object, multi-camera video surveillance systems received a lot of attention in recent years.
The view angleof 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 fewfisheyecameras can replace many traditional cameras to survey the events; such that the cost of system construction and management are then reduced. In this thesis, we propose an automatic detection and tracking system with two fisheye cameras for environment surveillance.The proposed system is composed of two major modules: foreground detection and foreground tracking.
In the foreground detection module,background subtraction is used to detect foreground pixels and logical morphology is exploited to connect foreground pixels as blobs and remove noises. Shadow areas are removed based on the characteristics of shadow that is a small block in background image with a significant change inintensity. Background update mechanism can adapt to the rapid and slow light change.
Foreground tracking is accompanied with Kalman filtering for pedestrian motion prediction. A transform table is pre-established to associate multi-camera data in the overlapping areas.When objectacross disjoint camera views, the data in the lookup table can provide enough information to realize the moving object in camera views actually belonging to the same object,and keep consistent labels on the object.To improve the reliability of the tracking performance, motion and color appearance features are used to match the detected objects in different cameras.Every camera has its own trackerto trace multiple target trajectories even if the moving objects are partial and complete occluded.
We conducted experiments with the proposed system on several videos; the environments of these captured videosare varied in brightness and have different object numbers. Theexperiments results show that the average sensitivity is 96.7 percent and the average false positive rate is 0.45 percent because the foreground objects are similar to the background. The average sensitivity rises to 98.55 percent with the Kalman filter.It demonstrates that the proposed method can work well under challenging conditions, such as light change, shadow interference, objectocclusion.So the proposed joint detection and tracking system is effective andreliable in practice. |
