電腦視覺方法應用於PC廣域即時車輛偵測與交通目標之追蹤

Real-timeArea-wide Vehicle Detection and Tracking by Computer Vision Methods with PC Architecture

即時偵測車輛在廣大範圍的運動狀態，一直是交通工程師極欲獲得卻尚未實現的理想。這個理想境界己可在本研究攝影所及的空間中，藉由即時面式車輛偵測來達成。影像「點」與「線」的偵測方法，對於交通偵測資料的空間連續性有所漏失，故無法暸解整個交通「面」的訊息。唯有其備大範圍即時追蹤車流運動軌跡的偵測能力，對於路面上的擁塞、交通事件、轉向、變換車道等交通狀況與駕駛行為才能立即而充分的掌握。本研究於PC上以電腦視覺方法完成廣域即時車輛偵測與交通目標之追蹤。本研究發展具有RST不變量的全面式廣域即時交通目標追蹤系統HAWK，可即時鎖定車輛各種運動方向的位置，未來可望提供交通工程師更完全的空間連續性資料，有助於即時性交通控制與評估。

This study establishes a real-time areawide vehicle detection and tracking system using computer vision technologies. Correlation coefficient matching method is adopted for image matching. The concept of object linking box is used to identify the characteristics of a vehicle. By comparing object linking boxes of two consecutive images, a specific vehicle can be locked and tracked along a roadway. The processing time of each image is 1/30 second with a deliberately designed personal computer system. Real-time detection can be achieved To apply the proposed system to traffic incident detection, a new algorithm is developed by examining time and location of each vehicle. Finally a case study is conducted to demonstrate potential applications of this system.