影像處理與粒子濾波器於智慧型車輛偵測系統之研究

Abstract

由於人類社會的演進，除了科技的進步之外，人們對於科技的需求也越來越多樣化，智慧型運輸系統已經成為現今研究最重要的議題之一。在現今智慧型運輸系統之應用之中，以影像為基礎之車輛偵測技術的需求越來越殷切，其中又以交通流量監控以及側邊盲點車輛偵測之應用非常需要。因此，本論文分別針對交通監控與盲點車輛警示系統應用提出能克服環境壅塞、光線等問題的車輛偵測系統演算法。 在第二章中提出了針對壅塞車況的車輛偵測演算法。以影像為基礎的車輛偵測系統中，背景相減法是最不會受到物體外型變化影響且效率很高的方法。但是隨著真實世界的車流環境的限制，壅塞時候的車流情況也需要考慮進去，其中，在長時間被車輛遮住背景的情況下與維持良好背景模型品質上，背景更新扮演著非常重要的角色，因此，本論文提出一小範圍更新演算法(Small range updating)以克服在壅塞車流環境下的背景更新。同時考慮到壅塞車流的影響，我們同時提出混和多種資訊的模糊演算機制（Fuzzy Hybrid Information Inference Mechanism，FHIIM）來做追蹤車輛的驗證，利用取得的特徵進行模糊推算並保留錯誤補償的機制來提高正確率。 在第三章中，我們提出一個以粒子濾波器架構加上相對區別方向梯度直方圖(Relative Discriminative Histogram of Oriented Gradients, RDHOG)為基礎的影像式車輛追蹤系統。本文利用相對區別方向梯度直方圖描述追蹤車輛的特徵再輔以粒子濾波器計算各粒子與目標車輛之間的關聯度（權重），最後根據最大事後機率Maximum a Posteriori Probability (MAP)決定估測的目標位置。此外，雖然相對區別方向梯度直方圖可以提供目標物的輪廓線索，但是追蹤車輛在偵測區域內會有尺規變化(Scale)的問題，因此本文提出了一個二階段關聯度計算，每一個粒子皆有兩個樣本(Sample)，一個是保有原來寬高的樣本，一個是寬高乘上一變化因子的樣本。透過此架構，經過二階段關聯度計算可以突顯不同大小尺規的粒子與目標物之間的差異，最後當做估測目標大小的根據。同時在本文中，將車輛間交疊區域的方向梯度直方圖進行調整以降低交疊區域對追蹤車輛的影響。 在第四章中，我們提出一個可進行日間與夜間的嵌入式盲點車輛偵測系統，使用影像分析為基礎進行盲點區域車輛偵測與追蹤，首先對裝設在車上的攝影機做詳盡完整的分析與介紹，接著介紹對輸入影像的感興趣區域如何選取。而盲點偵測演算法將分成日間與夜間進行討論。在日間的情況下，系統將採用邊緣與陰影複合平面取目標車輛的底部陰影，以日間車底陰影做為依據來偵測盲點區域中的車輛，再使用水平與垂直邊緣來確認車輛，最後進行車輛追蹤以確認是在連續畫面中的同一輛車。在夜間的情況下，本系統對輸入影像進行明亮物件萃取，透過直方圖門檻值搜尋(Automatic Histogram Thresholding)演算法擷取明亮物件，然後車燈物件利用時間域與空間域的特徵進行確認以刪除干擾的路面反光與分隔島造成的影響。最後利用追蹤該車燈軌跡來判斷是否有車輛於盲點區域。除此之外，本系統在偵測到盲點區域出現車輛時，系統還能判斷是否為前進、後退還是相對靜止，當系統判定為前進與相對靜止時，系統會發出警示以輔助駕駛，避免發生側撞的情況。本文亦將此方法在嵌入式平台DM6437上實現，並實際在實驗車Taiwan iTS-II上在高快速公路上與一般道路實驗。最後，在第五章的部分，我們整理了本篇論文的結論與未來的研究展望。

Due to the evolution of the human society and recent advances in technology, a great diversity of requirement for the technology by human being increases. Therefore, Intelligent Transportation System (ITS) has become one of the important issues in the current studies. Among the researches of ITS, vision-based traffic monitoring, and Blind Spot Detection (BSD) techniques become more and more significant in the current days. Therefore, vehicle detection system is an essential and important module in these applications. Hence, in this dissertation, several vehicle detection algorithms are proposed to solve problems in congestion, lighting, different weather conditions. Chapter 2 presents a new approach to video-based traffic surveillance using a fuzzy hybrid information inference mechanism (FHIIM). Background subtraction which is not affected by the appearance variation greatly is a high efficiency method for vehicle detection. However, due to the limitation of the traffic environment in the real world, the congested traffic situation should be considered in the traffic monitoring. Consequently, background update procedure plays an important role to maintain a good background model when the road surface is covered by the vehicles for a long time. Therefore, a small range updating is presented to conquer the background updating problem in the congested condition. Besides, FHIIM is employed to verify the tracked vehicles. Finally, a method that compensates for error cases under congested conditions is applied to refine the tracking qualities. In Chapter 3, a relative discriminative histogram of oriented gradients-based particle filter (RDHOGPF) approach to traffic surveillance with occlusion handling is proposed. Based on the conventional histogram of oriented gradients (HOG), an extension known as relative discriminative histogram of oriented gradients is addressed, which enhances the descriptive ability of the central block and the surrounding blocks. RDHOGPF can be used to predict and update the positions of vehicles in continuous video sequences. RDHOG was integrated with the particle filter framework in order to improve the tracking robustness and accuracy. To resolve multi-object tracking problems, a partial occlusion handling approach is offered, based on the reduction of the particle weights within the occluded region. Using the proposed procedure, the predicted trajectory is closer to that of the real rigid body. RDHOGPF can track the target by using the feature descriptor correctly in low contrast and very bright situations. In Chapter 4, an effective vehicle and motorcycle detection system in the blind spot area in the daytime and nighttime scenes is presented. The proposed method identifies vehicle and motorcycle by detecting and locating the shadow and the edge features in the daytime. Moreover, the vehicle and motorcycle could be detected through locating the headlights at nighttime. The bright objects in the nighttime are extracted based on automatic histogram thresholding to avoid a fixed threshold problem. The lamp feature, against the reflection of the pavement, is verified by spatial and temporal features. Finally, the driving behavior, such as relative approaching, relative static, and relative backing, are judged by the trajectory, respectively. Additionally, the proposed real-time vision-based Blind Spot Safety-Assistance System has implemented and evaluated on a TI DM6437 platform to perform the vehicle detection on real highway, expressways, and urban roadways, and works well on sunny, cloudy, and rainy conditions in daytime and nighttime. Experimental results demonstrate that the addressed vehicle detection approach is effective and feasible for vehicle detection and collision warning in various environments. Finally, we give a brief conclusion and future works in Chapter 5.