利用雙紅外線產生行人偵測系統中之影像待選區

Abstract

智慧型車輛已逐漸成為汽車工業中主要的趨勢，駕駛輔助系統在車輛安全、防止碰撞上更扮演一個重要的角色。駕駛輔助系統透過電腦與額外的感測器架構輔助駕駛，協助駕駛辨別遠方即將接近的障礙物與行人。 行人辨識系統的架構方面，通常分為3個階段，即分別是影像待選區域(ROI)產生、區域特徵偵測、行人追跡三個部份。 本論文即在探討配備近紅外線與紅外線的智慧型車載輔助系統中，利用影像邊緣特性，設法找出叢集特徵，並對於邊緣線段加以延展，判斷其特性後，產生影像待選區域。 實驗結果顯示，當偵測階段提高到一定的精準度時，運用此方式可以有效降低系統計算時間，成功迴避近紅外線與遠紅外線的不足，發展出更為有效率、高可靠度的方法。

Intelligent vehicles have gradually become the main trend in the automotive industry. Driver assistance systems, part of an intelligent vehicle, play an important role in safety and anti-collision domain. It helps drivers to identify the distance of the pedestrians and obstacles close-to-be by the computer and the sensor architecture. Pedestrian recognition system framework is usually divided into three stages, respectively. First is the Region of interest (ROI) generation; the second is regional detection; and the last is pedestrian tracking. This thesis is to study the intelligent vehicle auxiliary systems with near-infrared (NIR) and far-infrared (FIR) camera, focusing in the stage of ROI generation. We discover a method to find out the extending edges of existing cluster ones. So that , we could determine the feature and generate ROIs. Experiment results show that this method can decrease the computation time on a system with more accurate detection stage. The method successfully compensates the lacks from NIR and FIR, enhancing the system to be more efficient and more reliable.