利用子類別模型之影像暨距離資訊融合於多車道車輛偵測

Image and Range Sensor Fusion for Multilane Vehicle Detection Using Subcategory Model

林君丞(Chun-Cheng Lin)

指導教授 ： 傅立成

共同指導教授 ： 蕭培墉

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.10938

摘要

近年來，在智慧型車輛系統的發展層見疊出。智慧型車輛結合數種不同的感知技術，都是為了要了解車輛周遭環境的狀況，進而提供安全足夠的道路資訊給予駕駛者，甚至是達到自動化的駕駛服務。這些功能都倚賴於系統對於環境的感知，大至城市道路壅塞狀況，用以安排路徑，小至車輛周圍感測，提供路況資訊。在路況感測部分，路上障礙物感測尤其重要，路上障礙物的種類以車輛、機踏車、以及行人為多數。感測出路上障礙物，分析其運動行為，辨別障礙物是否危害駕駛者，便是邁向自動化駕駛服務的重大里程碑。
車輛影像偵測在近年來發展完善，然而考慮到其應用範疇，不同環境下的限制屢見不鮮。其中前方多車道的車輛偵測，因駕駛車輛與前方車輛相對位置變化差異，造成觀察前方車輛在外型上有著巨大的變化。若使用單一模型之影像分類器，外型上巨大的變化的組內差異很容易造成其效果不彰。
為了達到處理此一情形，本論文提出了影像暨距離資訊感知融合之多車道車輛偵測系統。在影像以及距離資訊上，將以單一類別多次模型分類器來個別處理因觀察角度造成外型巨大變化的情形。在影像方面，利用方向梯度直方圖特徵，配合多次模型串聯式分類器，處理在影像上的組內差異。在距離資訊上，以Inscribe Rectangle Filter Segmentation強健叢集之結果，並利用And-Or Model 描述車輛外型之特徵，分析觀察角度造成之特徵差異與空間關係。在感知融合上，藉由影像與距離資訊分類器得到之多子模型之結果，分析其機率模型建構之兩訊號源之間的關係，在多車道環境下依然有穩定的偵測效果。

關鍵字

多車道汽車偵測 ； 感知融合 ； 子模型分類 ； 叢集

並列摘要

In Taiwan, thousands of drivers and passengers injured in car accidents on the road each year. According to the statistics from Department of Statistic, Ministry of the Interior of Republic of China, there are 2612 people in sedan got killed in A1 class traffic accidents in Taiwan in last five years, with most fatal crashes involving more than one vehicle. Most of the causes of accidents are the unawareness of the driver to take care of the frontal driving condition on the road. It is critical to detect vehicles on the road for the safety of both drivers and other road users.
Image and range sensor fusion system for detection of vehicles has been a topic of great interest to researchers in the past decades. This fusion system can benefit from both image and range sensors. From the range sensor, it can retrieve the absolute information about the emergence of any obstacle only with minor noise. On the other hand, the ability of recognizing objects with particular hypothesis or specific feature is enabled to classify the object of interest from the help of image sensor. But it is still very difficult to verify the hypothesis for a computer to understand the environment like human beings do.
When it goes to the ability of descriptor, models in subcategory has been proved to be a better model than an overall model. As a result, we propose a sensor fusion system using subcategory model to achieve a robust detection system. In addition, there are new challenge when applying the subcategories into range information, and the Inscribe Rectangle Filter Segmentation is proposed to handle that challenge and And-Or Model is applied to describe the pattern extracted from the fine segmentation.

並列關鍵字

Vehicle Detection ； Sensor Fusion ； Subcategory ； Segmentation of Range Data

參考文獻

[1] S. Sivaraman and M. M. Trivedi, "Integrated Lane and Vehicle Detection, Localization, and Tracking: A Synergistic Approach," IEEE Transactions on Intelligent Transportation Systems, vol. 14, pp. 906-917, 2013.

[2] Y. Quan, A. Thangali, V. Ablavsky, and S. Sclaroff, "Learning a Family of Detectors via Multiplicative Kernels," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 33, pp. 514-530, 2011.

[3] H. T. Niknejad, A. Takeuchi, S. Mita, and D. McAllester, "On-Road Multivehicle Tracking Using Deformable Object Model and Particle Filter With Improved Likelihood Estimation," IEEE Transactions on Intelligent Transportation Systems, vol. 13, pp. 748-758, 2012.

[4] B.-F. Lin, Y.-M. Chan, L.-C. Fu, P.-Y. Hsiao, L.-A. Chuang, S.-S. Huang, et al., "Integrating Appearance and Edge Features for Sedan Vehicle Detection in the Blind-Spot Area," IEEE Transactions on Intelligent Transportation Systems, vol. 13, pp. 737-747, 2012.

[5] E. Ohn-Bar and M. M. Trivedi, "Learning to Detect Vehicles by Clustering Appearance Patterns," IEEE Transactions on Intelligent Transportation Systems, pp. 1-11, 2015.