漸進式交通標誌辨識系統

An Incremental Method for Developing a Road Sign Recognition System

本文主要目的在發展一套以人類視覺系統的形態辨識過程為基礎的漸進式交通標誌辨識系統，當系統接收到車載型攝影機拍攝到的連續影像時，交通標誌偵測模組立即被啟動，以模擬人類透過注意力集中的方式，很快且正確地偵測出影像中交通標誌的所在位置並分辨出它的種類（如紅色禁止標誌、紅色警告標誌等），然後交給辨識模組進行辨識的工作。但是，交通標誌偵測模組偵測出影像中的交通標誌時，此時交通標誌可能因距離車子還很遠，導致影像特徵不足而使交通標誌辨識模組無法做出最正確的辨識。因此，本系統先以僅有的物件特徵對交通標誌做粗略的猜測，同時系統繼續搜集更多的影像特徵，不斷地修正錯誤的猜測，直到辨識出交通標誌的類別為止。實驗結果證實這個辨識方法不但具可行性及擴充性，亦可套用在其它辨識方面的應用。

Rpad sign recognition is an important task in driver assistance systems. In this paper, we propose an incremental recognition approach based on the process of pattern recognition of the human vision system. When features of an object are detected, the human brain will make some reasonable assumptions based on these features and previous related experiences, and the vision system will collect more advanced features to verify these assumptions at the same time. if all features of the object are consistent with the viewer&apos；s cognition of it, then he "recognizes" the object. If he finds his assumptions to be incorrect he will make more reasonable ones, and will continue to find more features to support these new assumptions until he recognizes the object. Based on such incremental recognition procedures, then, we have developed an incremental road sign recognition system. When the system receives the video sequence captured by an on-board camera, the road sign detection module is activated and the road sign can be located and roughly classified based on the human mode of successive visual focusing； then the task is transferred to the road sign recognition module. However, the features of the road sign may be not sufficiently distinct for its to be correctly identified； for instance, the sign may be too far from the car. Therefore the system makes some assumptions according to the features it can detect and, in the meantime, it continues to collect more advanced features to modify its incorrect assumptions until the road sign is recognized. Our experimental results show that his system is reliable and feasible, that is, easily transferable to other recognition systems.