適應性前方物件偵測演算法應用於嵌入式即時影像輔助駕駛系統

A Vision-Based Adaptive Front Object Detection Algorithm Applied to an Embedded Driving Assistant System

葉展昕(Chan-Hsin Yeh)

指導教授 ： 曾百由

國立臺北科技大學/機電學院/機電整合研究所/碩士(2013年)

摘要

一般影像辨識系統都是建立在高速運算的電腦平台上，但為了開發應用於智慧型車輛系統上之影像系統，本研究採用嵌入式數位訊號處理器，使成本、耗能與體積上達到縮減，以符合車用電子之需求。本嵌入式系統選用Analog Devices的ADSP-BF561數位訊號處理器作為本研究的演算法開發核心，使用專為影像設計的硬體週邊功能，配合內部韌體設定使系統整體處理效率得到優化，並使用該型號之開發套件完成開發。
本研究之車道線辨識演算法藉由Otsu二值化有利於分離對比強烈之物件的特性，將車道線與路面分離，再藉低成本的空間特徵演算法來偵測邊緣，配合車道線的特定型態與合理性，達到濾除雜訊的效果。前車辨識演算法藉由車體之空間特徵以及物體陰影邊緣在路面上的特性，以偵測前方車輛，並完成可辨識複數車輛的方法。本研究之演算法在PC上開發，並成功的移植到嵌入式影像系統中，其中所開發之車道線偏離警示可在不同的天候環境下執行，輔助駕駛者做出適當的行車判斷；前車碰撞警示則包含側邊物件判斷，提醒駕駛者前方及側方路況。本研究使用嵌入式平台開發的輔助駕駛系統演算法，對於本系統頻率約每33毫秒更新一次畫面而言，演算法總處理時間在容許範圍內，足以達到即時辨識的目的，完成一個可實際應用於智慧型車輛之嵌入式影像系統。

關鍵字

嵌入式系統 ； 數位訊號處理器 ； 車道線偏離警示 ； 空間特徵演算法

並列摘要

Vision-Recognition systems are generally built on the high-speed personal computer(PC) platform. However, in order to be integrated in intelligent vehicle systems, this study uses the embedded Digital Signal Processors (DSP) so that the cost, power consumption and volume can be reduced to comply with the specification of automotive electronics. This embedded system selected the Analog Device’s ADSP-BF561 digital signal processor as the core of the development and the hardware peripherals designed specifically for image process with internal firmware settings to optimize the overall system processing efficiency. A development kit has been used to complete of this study.
In this study, the lane mark detection algorithm was improved by Otsu’s method, which is conducive to separate characteristics of objects will strong contrast features to find out lane mark, and to detect the edge with low cost knowledge-based algorithm. As the noise can be filtered, the lane marks with particular and reasonable can be identified formats. To achieve front vehicle detection, the algorithm using the spatial characteristics of the vehicle body and characteristics of shadow edges on the road to detect the front vehicle. The developed algorithm is capable of identify multiple vehicle. The algorithm has been developed on a PC, and ported successfully to embedded image systems. The lane departure warning can be performed under different weather environment, to make appropriate decisions helping the driver; Front car collision warning including side objects detection, can alert the driver for the front and side conditions on the road. The driving assistance system algorithms developed on embedded platform used in this study, can run at an update rate of 30Hz. The total processing time is within the allowable range sufficient to meet the requirement of real-time detection, for a embedded image systems applied in intelligent vehicle.

並列關鍵字

Embedded systems ； digital signal processor ； lane departure warning ； knowledge-based algorithm

參考文獻

6. STMicroelectronics

7. Rene Sosa, Gerardo Velazquez, “Obstacles Detection and Collision Avoidance System Developed with Virtual Models,” IEEE International Conference on ITESM (ICVES), Dec. 2007.

8. 蕭任斌，分散式救援機器人之即時影像防震與伺服定位，碩士論文，國立交通大學電機與控制工程學系，新竹，2009。

9. Jon Arr’ospide, Luis Salgado, Marcos Nieto, and Fernando Jaureguizar, “On-board Robust Vehicle Detection and Tracking Using Adaptive Quality Evaluation,” IEEE International Conference on Image Processing (ICIP), 2008, pp. 2008-2011

10. JongBae Kim, “A Real-time Moving Object Detection using Wavelet-based Neural Network, ” International Conference on Consumer Electronics( ICCE) Digest of Technical Papers, Seoul Digital Univ., Seoul, Jan 2009.