車聯網之應用:基於點集拓樸學比對及kNN最近鄰居法之車輛合作定位系統

An Application of V2V Communication: Cooperative Vehicle Positioning System based on Topology Matching and k-Nearest Neighbor Algorithm

張仲霆(Chung-Ting Chang)

指導教授 ： 連豊力

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201602643

摘要

近年來，隨著汽車對汽車通訊領域(汽車對汽車通訊即汽車之間能透過無線通訊裝置分享信息)及汽車對基礎設施通訊領域(汽車對基礎設施通訊即汽車與基礎設施之間能過無線通訊裝置分享信息)之研究日益完善，先進輔助駕駛系統將分為自給自足系統以及互助系統。自給自足系統經常受限於光的直進性，造成無法順利被遮蔽之車輛，而互助系統能透過周遭車輛所傳送的空間資訊來預測被遮蔽車輛之位置，進而避免汽車撞擊前方車輛。
本論文假設所有車輛都能透過測距儀與GPS感測器產生含有對周遭車輛之位置估測的局部道路地圖，並且能夠過無線通訊裝置廣播局部道路地圖給周遭車輛。當自走車接收到許多周遭車輛所發送的局部道路地圖時，自走車先透過點集拓樸學比對將接收的局部道路地圖與自走車所產生的局部道路地圖做匹配。接著透過基於最近鄰居法的自動分群演算法將這些屬於同一台車的不同量測結果作自動分群。最後，根據感測器的精確度，將分群後的量測點以適應性位置估測來計算該車輛的位置。
本論文提出車輛合作定位系統之模擬結果以及實驗結果。模擬結果呈現出所提出的車輛合作定位系統能偵測到的車輛數，在大部分的時間都比單靠一個測距儀偵測周遭的車輛還來的多，這也代表提出的系統能讓車輛對環境的認知有更寬廣的視野來加強道路安全。實驗以立體攝影機作為車輛上的測距儀，並在真實場景中量測周遭車輛的位置。在場景中，自走車的附近有三台車。首先，自走車僅用立體攝影機對周遭車輛作測距，實驗結果呈現出立體攝影機對於中間車輛的測距精準度比對於兩側車輛測距的精準度還要好。第二，自走車以所提出的車輛合作定位系統對附近的三台車輛做測距，自走車本身的位置由4個量測點來估測，其中一個量測點為自走車本身量測到的GPS位置，另外三個量測點分別為另外3台車透過各自的GPS位置再加上各自對於自走車所量測的相對距離位置而得到。實驗結果呈現出所提出的系統之平均測距精準度比僅用立體攝影機之平均測距精準度還更好。

關鍵字

汽車對汽車通訊 ； 先進輔助駕駛系統 ； 前方車輛防撞 ； 點集拓樸學比對 ； 最近鄰居法 ； 測距 ； 車輛合作定位 ； 立體攝影機

並列摘要

As V2V (Vehicle to Vehicle) and V2I (Vehicle to Infrastructure) area are well researched in recent years where the V2V technology can allow vehicles share information with nearby vehicles and the V2I technology can allow vehicles share information with nearby infrastructures by wireless communication device. The advanced driving assistance system can be divided into self-sufficient systems and interactive systems. The interactive systems, as the name implies, interact with infrastructures and/or other vehicles where these systems receive spatial information from nearby vehicles to prevent from forward collision. While the self-sufficient systems are limited to line-of-sight detection, the interactive systems account for scenarios farther ahead by predicting the position of occluded vehicle.
In this thesis, each vehicle is assumed to generate a local map which is a set of position measurements of nearby vehicles by using onboard low-cost GPS and ranging sensor, and shares it with the nearby vehicles by broadcasting via wireless communication device. When the ego-vehicle receives multiple local maps from nearby vehicles, the received local maps are matched with the local map generated by ego-vehicle by topology matching. The position measurements belong to the same vehicle are clustered by automatic points clustering based on k-Nearest Neighbor algorithm. Those position measurements belong to the same vehicle are combined by adaptive position estimation which updates position estimation according to accuracy of the sensor currently.
In this thesis, both simulation results and experimental results by proposed cooperative vehicle positioning system are presented. The simulation results show that the number of detected vehicle by the proposed cooperative vehicle positioning system is more than by a single sensor alone in most of the time. It turns out that a vehicle can get an extended view of surroundings to improve driving safety. The stereo camera is used as a ranging sensor equipped on the vehicle to produce position measurements in a real scenario. In the scenario, there are 3 vehicles nearby the ego-vehicle. First, the ego-vehicle estimates the range to the other 3 vehicles by stereo camera only. The experimental result show that the stereo camera gets a higher range estimation accuracy to the middle vehicle than the side vehicle. Second, the ego-vehicle estimates the range to the other 3 vehicles by the proposed cooperative vehicle positioning system. The position of the ego-vehicle is estimated by 4 measurements where 1 measurement is measured by GPS sensor of the ego-vehicle and the other 3 measurements are measured by both GPS sensors and ranging sensors of the other 3 vehicles respectively. The experimental results show that the accuracy of range estimation by the proposed system is better than by the stereo camera only.

並列關鍵字

V2V ； advanced driving assistance system ； forward collision preventing ； topology matching ； kNN algorithm ； range estimation ； cooperative vehicle positioning ； stereo camera

參考文獻

[17: Aly 2008]

[1: Chavez-Garcia and Aycard 2016]

R. O. Chavez-Garcia and O. Aycard, "Multiple Sensor Fusion and Classification for Moving Object Detection and Tracking," IEEE Transactions on Intelligent Transportation Systems, Vol. 17, No. 2, pp. 525-534, Feb. 2016.

[2: Sivaraman and Trivedi 2013]

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, No. 2, pp. 906-917, June 2013.