Cooperative Dual-Camera System for Real-Time Target Searching and Close-up Image Acquisition

The main function of visual surveillance systems is to detect and track objects of interest. If a system can provide clear close-up images, it is very effective for target recognition, crime prevention and detection, and ensuring safety. The proposed architecture for a cooperative dual-camera system can effectively position and capture close-up images of targets moving in different directions. As such, it avoids any blind regions, which is a difficulty typically encountered in traditional master-slave surveillance systems. Employing human-body-structure analysis can reduce the effect caused by variations in illumination and shadows in an environment, and hence improve object-extraction outputs and generate fewer errors in the three-dimensional positioning of targets. Using the cross-ratio of geometric invariants to estimate a target's moving distance can enable the prediction of target positions within the required operational time of the cooperative system. In this way, a close-up camera can correctly capture close-up images of targets in real time. The proposed system employs pan-tilt-zoom (PTZ) cameras and determines the appropriate shooting distance, zoom, and focus to obtain clear close-up images of targets. Experimental results show that the proposed system can produce good close-up images of both single and multiple targets.

關鍵字

visual surveillance system ； cooperative dual-camera system ； target detection ； close-up image of the target

並列摘要

視覺監視系統除了要能完成目標物的偵測與追蹤外，提供清楚的特寫影像更有助於目標物辨識、犯罪偵防、與安全維護。使用協同式的雙攝影機系統架構能夠有效的針對不同方向行進的目標物，進行定位與特寫影像拍攝，避免了主僕式監視系統的拍攝死角問題。針對人體的結構進行分析，可以抵抗因為環境光影變化，而產生正確的目標擷取，進而使目標物三維定位的誤差下降。使用幾何不變量交比值的目標物運動量估測，可以配成協同式系統的運作時間要求，進行目標物空間位置的預估，使特寫攝影機能即時的拍到正確成像的目標物特寫。系統採用PTZ攝影機，同時考慮拍攝距離、zooming及focusing參數的配合，達到拍得清晰目標物特寫影像的目的。實驗顯示，所提出的系統能針對單目標與多目標產生好的特寫影像。

並列關鍵字

視覺監視系統；協同式雙攝影機系統；目標物偵測；目標物特寫

主題瀏覽