基於視覺停車格偵測與多階段控制之自動停車系統開發

Development of Auto Parking System with Vision-based Parking Space Detection System and Multi-steps Control System

李則霖 ； 詹聖瑋

《機械工業雜誌》 433期 (2019/04) Pp. 79-86

摘要

隨著當前汽車無人駕駛技術日新月異的發展，國內外車廠紛紛投入汽車無人駕駛技術的研發，其中，自動停車技術為當前重要發展標的之一。而現有半自動停車系統，多採用距離感測器偵測停車空間，需要依賴鄰車以定位停車空間，造成使用者在僅有地上標線的停車格進行停車且該停車格兩旁無鄰車可協助偵測時，反而造成停車不易的現象。本文利用基於視覺的停車格偵測系統，用於即時處理和識別車格線，以提供車輛和格線之間的相對空間資訊。使用基於視覺的算法不需依賴周遭參考障礙物。我們還將視覺資訊與其他感測資訊整合在一起，以改善性能，分析環境和車輛狀態後，制定適當的自動停車規劃。然後，處理器控制車輛的致動器，使自動駕駛車輛能完成垂直，平行和斜向車格的停車任務。因此，本文解決了在沒有參考障礙的情況下在車格邊線區域停車的問題。

關鍵字

視覺偵測 ； 機器學習 ； 自動停車

並列摘要

With the rapid development of current autonomous driving technology, domestic and international car manufacturers have invested in development of autonomous driving technologies, among which, automatic parking technology is one of the current important development targets. In the existing semi-automatic parking system, the distance sensor is often used to detect the parking space, and neighboring vehicles are required to locate the parking space, so that it's not easy for users to park in a parking space without neighboring vehicles but only with the ground markings. This article is based on the vision-based parking detection system for instant processing and identification of grid lines to provide relative space information between vehicles and grid. Using vision-based algorithm wouldn't be limited by neighboring vehicles. We also integrate visual information with other sensing information to improve the performance, and help determine appropriate automatic parking plan by analyzing environment and vehicle status. The processor then takes control of the vehicle's actuators, making auto vehicles to complete parking missions for perpendicular, parallel, and oblique grids. As a result, this study solves the issue of parking in a grid-painted region without reference obstacles.

並列關鍵字

Vision-based detection ； Machine learning ； Auto parking