在嵌入式異質雙核心平台上整合開發夜間車燈與車道追蹤偵測與事件記錄功能之駕駛輔助系統

Integrating and Developing the Night Lights and Lane Tracking Detection and Recording Event Functions into the Driver Assistance Systems on the Heterogeneous Dual-Core Embedded Platform

魏仟豪(Chien-Hao Wei)

指導教授 ： 陳彥霖

國立臺北科技大學/電資學院/資訊工程系研究所/碩士(2012年)

https://doi.org/10.6841/NTUT.2012.00514

摘要

由於全世界每年平均有超過一千萬人傷亡是肇因於車禍意外，而其中有半數的交通意外又是發生在夜晚，因此本論文的目的是以電腦視覺技術為基礎發展一套嵌入式夜間駕駛安全輔助系統，並藉由架設於汽車擋風玻璃後方之視訊擷取裝置來即時截取前方的影像，再經由一系列的電腦視覺技術加以分析處理來提供夜間行車輔助時所需的車輛和車道線辨識與追蹤、事件偵測和即時行車記錄等個模組所需之資訊。
在車輛偵測上使用影像切割與型態分析技術來標定出車輛的車頭燈與車尾燈，而在車道線偵測上則利用車道線特徵模組來取出此車道線特徵點在畫面中的所在位置，並將所得到的所有位置利用雲形曲線來建立車道線，再利用上述的位置資訊來進行事件判斷並使用H.264壓縮技術來即時記錄行車影像，最後整合並實現於一套異質雙核心(ARM+DSP)之嵌入式系統平台上。在實作上會先整合車輛偵測模組、車道線偵測模組、事件判斷模組和行車記錄模組，再依照DSP的特性對這些模組做一定程度的優化，使其系統能再異質雙核心之嵌入式平台上達到最佳分工效果，以實現一套應用電腦視覺技術的夜間嵌入式駕駛輔助系統。

關鍵字

電腦視覺技術 ； 異質雙核心 ； 嵌入式系統 ； 駕駛輔助系統 ； 事件偵測 ； 車輛辨識 ； 車道線辨識

並列摘要

The car accidents have been one of the most serious problems. There are over ten million casualties in average each year due to the car accidents, especially in nighttime. Therefore, this thesis proposes a nighttime embedded driving assistance system based on computer vision technology. By analyzing the captured images, the proposed system is able to faciliate vehicle detection, lane detection, lane tracking, event detection and real-time event recording mechanisms.
To detect vehicles based on features of vehicle headlights and taillights, the techniques of image segmentation and pattern analysis is proposed in this study. For obtaining lane features based on lane patterns and spline curves, the lane features and moving vehicle locations are applied for determining the possible traffic events and then activating the real-time video recording process based on H.264 compression technology. The utilized computer vision technology in this thesis is integrated and implemented on the heterogeneous dual-core embedded platform. For optimizing the algorithms, the vehicle detection, lane detection, event identification and video recording modules are integrated on an embedded platform by DSP platform-oriented optimization libraries. Finally, all the modules are integrated in an embedded system to achieve an intelligent embedded night driver assistance system.

並列關鍵字

computer vision technology ； heterogeneous dual-core ； embedded system ； driver assistance systems ； event detection ； vehicle detection ； lane detection

參考文獻

[25] 林敬俊，整合夜間車燈與車道偵測與事件記錄功能之嵌入式駕駛輔助系統，碩士論文，國立臺北科技大學資訊工程學系，臺北，2011

[1] D. Pomerleau, “RALPH: Rapidly adapting lateral position handling,” in Proc. IEEE Intelligent Vehicle Symp., pp. 506–511, 1995.

[2] D. Langer and T. Jochem, “Fusing Radar and Vision for Detecting, Classifying and Avoiding Roadway Obstacles,” in Proc. IEEE Intelligent Vehicle Symp., pp. 333 - 338, 1996.

[3] M. Bertozzi, A. Broggi, and A. Fascioli, “Vision-based intelligent vehicles: State of the art and perspectives,” Robot. Automat. Syst., vol. 32, pp. 1–16, 2000.

[6] C. J. Taylor, J. Malik, and J. Weber , “A Real-time approach to stereopsis and lane-finding”, in Proc. 1996 IEEE Intelligent Vehicles Symposium, pp. 207–213, 1996.