A Vision-Based Traffic Light Detection System at Intersections

Abstract

十字路口紅綠燈自動偵測系統為發展以視覺為基礎之道路交通違規偵測系統之重要基礎，例如闖紅燈違規、紅燈違規左右轉或暫停於禁止臨時停車位置等違規都需要紅綠燈的資訊。這類型的視覺系統由於是架設於室外開放的環境，因此必須要能夠克服天氣、陰影、光線明亮度變化以及架設位置不固定等相關的問題。我們希望能夠研究出一個以視覺為基礎之十字路口紅綠燈自動偵測系統，可以提供紅綠燈的訊息資料，而不需與紅綠燈控制信號箱有實體的連線，本系統所偵測的紅綠燈訊號即可配合其他智慧型交通運輸系統(ITS)做後續處理。本系統首先產生背景影像，並藉以判斷明亮度值—分辨白天或黑夜。其次採用HSI彩色模式配合模糊Morphology技術處理來取得可能的紅綠燈位置。配合空間與時間上的限制條件，將不合理的候選區域去除，即可將紅綠燈的位置、大小與轉換時間偵測出來。

The traffic light detection system is one of the key components of the vision traffic law enforcement system, such as red light runner detecting, turning against traffic light, and stopping at the non-stopping zone. With various conditions of both open outdoor environments and device setups, the traffic light detection must be robust to weather and illumination conditions, and also tolerant to various perspective angles.An automatic traffic light detection system at intersections is presented in this paper. It performs traffic light detection on traffic videos without any signals from the traffic light controllers. This system is useful to be integrated with another ITS (Intelligent Transportation System) components. Background images are first generated by the system and in the mean time illumination parameters are estimated. The HSI color model is employed, and fuzzy methods together with morphological technique are utilized to acquire the candidate traffic light areas. With the relative spatial and temporal information, the scales, positions, and timing sequences of traffic lights are obtained. Some results from a preliminary trial are reported, and the associated researches are in progress.