隨著駕駛安全意識日益增強，交通運輸行業最近致力於開發先進駕駛輔助系統(ADAS)以幫助駕駛員觀察車輛周圍環境條件的變化。單眼攝影機更是ADAS中捕捉車輛周圍信息的重要感測器。因此，基於視覺的技術是偵測障礙的主要方法。本研究採用夜間車輛檢測方法，利用RCCC感測器取代傳統拜耳濾色器(Bayer Filter)，重點在期望獲得更好的夜間圖像品質，候選的測試區域選擇基於時空分析、並採用三種不同的全局驗證來減少錯誤檢測結果。在檢測過程中利用了新的自適應跟踪檢測方法(Struck: Structured Output Tracking with Kernels)來增加系統的穩定性。
Abstract— With the consciousness of driving safety growing increasingly, traffic and transportation industry have been devoted to developing Advanced Driver Assistance System (ADAS) recently to help drivers observing the variation of the environmental conditions around the vehicle. Monocular camera is the essential sensor in ADAS to capture the information around the vehicle. Therefore, vision based technology is the main methods to detect obstacles. ADB systems typically exploit a camera that detect the front road view, if the system detect preceding vehicles or oncoming vehicles, the distribution of beam will changed to avoid glare other driver’s eyes. Consequently, vehicle detection at nighttime condition is the critical part of an ADB system. This study implements a nighttime vehicle detection method exploiting a RCCC sensor to replace Bayer sensor and aim to obtain a better quality of images at night. Candidate region selection is based on spatiotemporal analysis and three global verifications are adopted to reduce the false detecting results. A new adaptive tracking by-detection framework based on structured output prediction is applied with the detection process.
Keywords—ADAS, RCCC sensor, Vehicle detection, Spatiotemporal analysis, Tracking
- IA-06-0018 (482K)