交通背景型態分類模式之研究

A Traffic Background Pattern Classification Model

陳昶志(Chang-Chih Chen)

指導教授 ： 范俊海

淡江大學/管理學院/運輸管理學系碩士班/碩士(2007年)

https://doi.org/10.6846/TKU.2007.00614

摘要

自西元1970年代後，美國外學者Schlutsmeyer的研究證實了影像處理應用於交通工程上之可行性後，近年來影像式偵測器已逐漸蔚為主流。但影像式偵測器在實際的應用上受到交通環境背景複雜、照明度不足，以及光源不穩定等相關背景因素造成偵測效果不佳，先前的相關研究大都以白天且晴朗的路段或夜間晴朗的路段作為研究環境，對於環境之光線變化與歸類鮮少有探討，因此本研究以一般的郊區路段作為研究的環境，以期構建一個適用於交通環境分類之模式。
交通環境分類的良窳取決於光線的變化與色彩間的交互影響，以晴朗的下午為例，人類可輕易判斷出當下的天氣的狀況與選擇適當的影像偵測方法，但影像偵測系統卻需根據捕捉的畫面，分析影像中的色彩特性與分佈才能判斷天氣狀況。因此如何在交通環境中辨識出當下的天候與光線是本研究探討的重點。如何將影像中的車輛移除並建構出無車的交通背景，避免偵測與抓取色彩特徵受車輛干擾是本研究所要克服的第一個難題，本研究提出以中位數法(Median Filter)結合遞迴式(Recursive)更新之背景重建模式，用以獲得不受車輛干擾的背景影像。將不受干擾的背景影像建構完成後，觀察擷取出的交通背景影像色彩分佈，在不同的環境下色彩分佈亦不同，因此如何將色彩分佈的特性進行分類是本研究遭遇的第二個困難，對此本研究提出模糊類神經網路(Fuzzy-Neuron Network)與倒傳遞類神經網路(Back-Propagation Network)發展出一分類模式，依色彩間相關的特性與進行分類的工作，其研究結果顯示在不同的交通環境下色彩間具有近似的分佈，依據各個環境的特徵進行分類有不錯的效果。
本研究所偵測的交通環境包含晴朗的下午、黃昏與夜晚和雨天的下午、黃昏與夜晚。本研究將天候分類正確或光線分類正確定義為第一種正確率，將天候分類正確且光線分類正確定義為第二種正確率，經實例驗證後，模糊類神經網路於第一類正確率為98.11%、94.34%；在天候與光線同時分類正確沒有誤判，正確率為92.45%。

關鍵字

背景重建 ； 色彩模式 ； 類神經網路 ； 模糊系統 ； 分類

並列摘要

A good traffic surveillance system must be capable of working in all kinds of weather and illumination conditions. Using image detection machine usually does not effective because it be affected by weather, bright, and complex traffic background. If we can choose good detection algorithms for vehicle detection depend on weather and illumination change. This paper presents reconstruction of background model and classification model to solve the problem of detect algorithms transform.
The result of classified traffic background depends on interaction of illumination change and color model. We can judge what kinds of weather and illumination by using our eyes and ears to choose suitable image detection algorithms. But if machine want to choose fit detection algorithms, it have to depend on capture frame. How to judge conditions of illumination and give a suitable suggestion for detection is this paper kernel. First, this paper presents a recursive median filter background model to remove vehicles form frame. It can avoid interference form vehicles and illumination change, which affect detective area. Second, according to color distribution, which retrieves from detection frame, the paper presents fuzzy-neuron network to classify all kinds of weather and illumination conditions. The result displays that color have similar distribution at the similar traffic condition, and it will contribute to classification of traffic background pattern model.
The experimental place is in outdoors. The weather includes sunny day and rainy day, and the illumination change includes afternoon, nightfall, and night. According to our experimental analysis, the accurate rate of weather classified is 98.11 %,and illumination classified is 94.34 %.Then the accurate rate of weather and illumination is 92.45 %.

並列關鍵字

Background ； Color model ； Neuron network ； Fuzzy system ； Classification

參考文獻

2.Stauffer C, Grimson W. E. L., “Learning patterns of activity using real-time tracking,” IEEE Transactions on Pattern Analysis & Machine Intelligence, 2000. 22(8): p. 747-57.

3.S.-C. Cheung and C. Kamath, “Robust techniques for background subtraction in urban traffic video,” in roceedings of Video Communications and Image

Processing, SPIE Electronic Imaging, Jan 2004.

5.Bhandarkar, S.M., Luo, X, “Fast and Robust Background Updating for Real-time Traffic Surveillance and Monitoring,” Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05).

7.C. Stauffer and W. E. L. Grimson, “Adaptive background mixture models for real-time tracking,” in Proc. IEEE Conf Computer Vision and Pattern Recognition　CVPR’99, pp. 246–252, 1999.