智慧型道路影像學習系統應用於車輛自動駕駛之實作研究

The Application of Artificial Neural Networks on Vision-Based Steering System for Unmanned Ground Vehicles

李天明(Ten-Min Lee) ； 周建明(Chien-Ming Chou)

《明道學術論壇》 3卷2期 (2007/12) Pp. 13-26

https://doi.org/10.6953/MJ.200712.0013

摘要

本研究將傳統之一般電控車經機電整合之特殊設計研改爲可完全由電腦自主控制(Autonomous Control)之無人駕駛載具，再應用類神經網路(ANNs)學習法之多層感知器(MLP)理論，學習人類真實駕駛之模式來控制車輛之行進方向，達成機器學習及車輛自主控制之研究目標。
爲求縮短研發時程及實驗操作之方便性，本研究採用電腦模擬及小型機器人爲研究初期測試之工具，先行驗證研究理論之可行性。因此，研究初期之操作測試可完全不受天候之影響，而大幅縮短了控制程式之開發時程，亦降低了無人操控車接受實驗測試之磨耗。
本研究之成果，目前已可成功利用影像處理之技術來判別真實道路之偏向角度，進而讓無人操控車可依據道路之偏向狀況而自動調整行車方向。本研究亦成功運用類神經網路學習法，訓練車輛之視覺辨識學習系統能夠成功學習車輛前方道路之數位影像與轉向系統間之正確驅動關係，而達成車輛可依既定道路自主行駛之目的。

關鍵字

類神經網路 ； 無人操控車 ； 影像處理 ； 機器學習 ； 自主控制

並列摘要

This paper describes an integrated technique to transform a normal electrical automobile to an autonomous controlled vehicle. In order to demonstrate the possibility of machine learning in the application of unmanned ground vehicles, a multi-layer perceptron (MLP) using the theory of artificial neural network (ANN) was implemented to learn the driving pattern from human beings.
In the initial stage of this research, a four-wheel robot was used as the simulation tool to prove the algorithm of control programs. As robot experiments can reduce the worn-out and are beyond constrains of weather condition, therefore the time and cost were saved in this research.
The results of this research showed that the unmanned vehicle can autonomously control its moving directions according to the road image ahead through our image processing system. The results also demonstrated that the MLP can successfully learn the relationship between the road images and the steering signals.

並列關鍵字

Artificial Neural Networks ； Unmanned Vehicle ； Image Processing ； Machine Learning ； Autonomous Control