用於遍布式即時交通資訊與路面品質收集與散佈的智慧型手機探偵車技術

Abstract

智慧型運輸系統(Intelligent transportation systems, ITS)是一門跨領域的整合學門，其主要著重於改善交通管理的效率。在交通管理問題中，旅行時間不僅是一個重要的路徑規劃指標，也影響了許多的管理議題。在市區，旅行時間主要是花在通過號誌路口所需的時間，另一方面，道路平整度除了影響到駕駛行車的安全與車輛的耗損之外，對於政府來說也是人民對於市政滿意度的一項重要指標。因此，了解並觀察號誌路口的交通行為以及提供時時路面監測是相當重要的。車載網路(vehicular networks)所採用短距通訊(dedicated short range communications, DSRC)已成為ITS所採用的基礎網路的候選人之一。現今，智慧型手機搭載了多種感測器提供了一個良好的平台以實作探偵車的概念於ITS中。在本論文中，我們提出了一個以智慧型手機為基礎的探偵車系統以散佈及分析交通資訊。在本論文中，我們完成了四項研究已分別針對市區號誌路段交通分析、路面偵測、市區訊息的散佈以及區域性訊息散佈的問題，提出解決方案。 紅綠燈系統中的號誌長度與號誌變換時刻，也就是所謂的時相資訊(phase timing information, PTI)，在現代運輸系統中扮演了相當重要的腳色，影響車輛在號誌路口的行為。然而，此系統卻不是可輕易讓公眾取得。而使用眾源資料方法來探勘交通資訊，相對於傳統基礎建設方法，例如：感測迴圈、閉路監視器等，來說有較低的佈建與維修成本。因此，此方法逐漸成為一種新的趨勢。然而對於眾源資料方法，探勘技術與參與資料收集的穿透率為影響此方法成效的最重要的兩個議題。在第一個研究中，我們針對號誌路口提出一個基於震動波理論的交通資訊估算框架，以眾源資料的方法透過利用在號誌燈前車輛停止與行走的事件來解決此一問題。我們透過允許部分車輛參與PTI搜尋以正規化PTI問題。我們的框架方法從偵測停走事件開始，接著透過壓縮不同號誌周期所偵測到的停走事件於同一號誌周期透過震動波技巧來計算PTI。我們分別透過實際場測與模擬實驗來驗證所提出的解法。實驗結果顯示單純使用GPS接收器，我們的偵測方法可偵測出大部分的停走事件。因此，我們的方法可直接用於目前市面上大部分的智慧型手機。此外，即使在非常低的穿透率情境中，約使用1.2%的停走事件，我們的震動波方法仍可計算出號誌周期長與號誌變換時間，方均根的誤差分別為2.3秒與8秒。此程度的精確度可應用於多種運輸應用中。此外，我們的實驗結果顯示，震動波模型非常適合從眾源資料中萃取交通資訊，而摺疊的概念可有效地降低穿透率的需求。 在眾源應用中，眾源資料的品質對於後續系統層級的探勘流程具有決定性的影響。在第二個研究中，我們提出一個智慧探偵車系統以用於監測路面平整度。一個智慧探偵車構成元件包含一般車輛以及掛載在車上的智慧型手機，而手機上運行感測程式以客觀的檢測因撞擊到路面異常，例如：坑洞、減速墊，所造成的振動。我們提出的系統有多個特色。首先，為了提供動態設定SPC，我們研發出一個訊號處理機制從加速度資料萃取出垂直分量，這些萃取出的垂直分量將用於後續的震動偵測與道路平整度檢測。透過這個方法，所提出的系統提供了一個對於駕駛友善的環境，讓駕駛無需考慮複雜的裝設以及使用設定，而這種環境將有利於日後進行大規模系統佈建，讓駕駛者願意參與眾源資料收集。其二，基於阻尼模型，我們提出了路面異常指標以代表不同路面異常，而非受到車況影響。此指標可有利於後續在伺服器上系統層級的探勘流程。第三，我們實作了SPC系統雛形，並透過大量路面測試以驗證與評估所提出的系統效能。此外，我們實驗了基於DENCLUE的分群演算法從回報資料中來探勘出路面異常資訊，以展示各種可能由系統層級所獲得的潛在效益。 由於車輛的高密度與高行動力的特性，針對市區設計一個有效率的車載網路廣播協定是一個困難的挑戰。在第三個研究中，我們提出了一個廣播協定，稱為streetcast，以提供有效率的廣播服務。電子街道地圖用來輔助轉傳點的選擇，多播RTS (multicast RTS, Request-To-Send)機制採用來保護無線通訊以提供高可靠的傳輸機制。此外，一個適性訊息控制機制被提出降低訊息負載。我們提出的協定透過實際道路情境與車流的模擬驗證下，結果顯示我們的方法有較高的封包傳輸率，約1.25倍相較於傳統氾濫(flooding)機制、5倍相較於多點轉傳 (Multi-Point Relay, MPR)機制；較短的點到點延遲，約1/4相較於氾濫機制、約1/3相較於MPR機制；與較少的碰撞次數，約1/11相較於氾濫機制與1/2相較於MPR機制。 在第四個研究中，我們提出一個基於COPE的無線區域封包傳送協定，稱為非同步訊息散佈(Opportunistic ASynchronous Information dissemination, OASIS)。 OASIS不僅繼承了COPE的兩個特色：隨機監聽(opportunistic listening)與隨機編碼(opportunistic coding)，也提出一個新的特色：隨機訊息散佈(opportunistic information dissemination)，積極盡可能將多個封包訊息編碼至同一個封包中即便該封包不是傳送至對應的下一站。我們從COPE中觀察到兩件事情：利用條件機率來猜測一個節點的解碼機率會比單純使用兩點鏈結封包傳送機率來的好，以及基於傳送編碼封包的吞吐量期望值來做編碼決策以替代簡單的整體傳送機率方法。根據這些想法，我們進一步提出一個機率行編碼演算法。此外，我們指出封包暫存區的管理與封包交換機制對於實作的重要性並提出一個簡單的封包暫存區管理機制與可變封包表頭格式。我們透過模擬在不同網路流量的網狀網路以比較OASIS、COPE與單點傳播的封包傳送效率。實驗結果顯示OASIS在網路吞吐量有約1.5倍相較於傳統的單點傳播，網路吞吐量的改進相較於COPE有約1.2倍的提升。

Intelligent transportation systems (ITS), which integrate inter-disciplines, have emerged to make traffic management efficiently. Among various traffic management issues, travel time is not only an important evaluation metric for route plans, but also related to many management issues. In urban area, travel time is mainly dominated by the time passing signalized road segments. In addition, road surface roughness is also an important factor to not only driving safety and damages to vehicles, but also a metric to reflect satisfaction of citizens to governments. It is important to monitor and study traffic behaviors in signalized road segments and road surface roughness. Vehicular networks adopting dedicated short range communications (DSRC) is one of the preferred network design options for ITS. Nowadays, smartphones equipped with many sensors provide a good platform to bring the concept of probe car systems into ITS. In this thesis, we propose a crowdsourcing approach, smartphone-based probe car (SPC) system, for real-time traffic information. Four works are considered to deal with problems in reliable message dissemination in urban areas, efficiently message dissemination in local areas, traffic dynamic analysis in urban signalized road segments, and road pavement monitoring. The cycle lengths and signal transition time of traffic light systems (TLS), or known as the Phase Timing Information (PTI), play a key role in modern transportation systems. However, such information is not always available to the public. Crowdsourcing is a new trend for pervasively discovering traffic information due to its low deployment and maintenance cost as compared with traditional infrastructure based approaches, e.g., loop detectors and CCTV. Mining techniques and the penetration rate of participators in the discovery process are two major issues in such approaches. In the first work, we propose a crowdsourcing approach to solve this problem by exploiting the stop and go (SG) events of the vehicles on the roads happening in front of those target traffic lights. We formulate the PTI discovery problem by allowing only part of the vehicles participating in the discovery process. Our framework starts with discovering SG events, followed by collapsing these events over multiple signal cycles into one and calculating PTI information through a shockwave technique. We have verified our solutions by field trials and simulations. The results show even with GPS receivers, our detection algorithm can find most of the SG events. Therefore, our approach may be directly adopted by current smartphones. Further, as an example even with a low penetration rate of SG events around 1.2% over all SG events, our shockwave approach is able to compute the cycle length, signal transition time of a TLS with root mean square errors of 2.3 seconds and 8 seconds, respectively. This level of accuracy can be helpful in many transportation applications. In addition, our results also show that shockwave models are useful to extract traffic information from crowdsourced data, and the folding technique can effectively reduce the requirement on the penetration rate. In crowdsourcing applications, the quality of the crowdsourced data is decisive to the success of subsequent system-level mining processes. In the second work, we proposed a Smartphone Probe Car (SPC) system to monitor road pavement. An SPC is essentially an ordinary vehicle with a mounted smartphone that runs sensing programs to objectively assess bumping caused by road anomalies such as potholes and bumps. The proposed system has several features. Firstly, to allow dynamic forming of SPCs, we develop a signal processing heuristic for the extraction of the vertical acceleration components from the accelerometer readings (Upon which bumping detection and road surface anomaly assessment rely). By this mean, the proposed system provides a driver-friendly environment, requiring neither complicated installation nor driver-assisted training processes, and thus is possible to achieve hassle-free mass deployment such that drivers would be willing to participate in crowdsourcing. Secondly, based on the underdamped oscillation model, we propose a road anomaly indexing heuristic that is representable for road anomalies rather than vehicle conditions. This will later facilitate the system-level data mining processes in the servers. Thirdly, a prototype SPC system was implemented and extensive field tests were undertaken to verify the performance of our system framework. Further, we experimentally adopted a DENCLUE-like algorithm to mine road anomaly information from reported events to demonstrate any potential benefit from future investigation of data mining process at the system level. Due to the high density and high mobility of vehicles, it is a challenge task to design an efficient broadcast protocol for VANETs in urban areas. In the third work, we propose a broadcast protocol, named Streetcast, to provide efficient broadcast service. Street maps are used to assist the selection of relay nodes, and a multicast RTS (Request-To-Send) mechanism is adopted to protect wireless communications for providing high reliability. In addition, an adaptive beacon control heuristic is proposed to reduce beacon overheads. The proposed protocol is evaluated via simulations in a real roadmap scenario with real traffic flows. The simulation results show that the proposed broadcast protocol has a higher packet delivery ratio, about 1.25 times of the flooding based schemes and 5 times of the MPR (Multi-Point Relay) based scheme; shorter end-to-end latency, about one-fifth of the flooding based schemes and one-third of the MPR based scheme; and fewer collisions, about one-eleventh of the flooding based schemes and half of the MPR based scheme In the fourth work, a localized packet forwarding protocol based on COPE for wireless networks, called Opportunistic ASynchronous Information diSsemination (OASIS), is proposed. OASIS not only inherents two features from COPE, opportunistic listening and opportunistic coding, but also introduces a new one, opportunistic information dissemination, which aggressively encodes as many packets as possible even if packets are not going to be received by their next hops. We have two observations from COPE, i.e., applying conditional probability to guess decodable probability of a node is better instead of link delivery probability, and based on expected throughput of sending a coded packet for making coding decision instead of simple overall probability. Based on these ideas, we further present a probabilistic coding algorithm. In addition, we point out that packet pool managements and packet information exchanges are critical in the implementation and therefore propose a simple pool management mechanism and a variable header format. We evaluate OASIS with COPE and unicast via simulations under different traffic loads in mesh networks. Simulation results show that OASIS has network throughput about 1.5 times of traditional unicast forwarding, and the improvement is about 1.2 times of the improvement achieved by COPE.