具有情景感知的智能汽車：從模型到原型的發(fā)展畢業(yè)論文外文翻譯

1、Context-aware smart car: from model to prototypeAbstract: Smart cars are promising application domain for ubiquitous computing. Context-awareness is the key feature of a smart car for safer and easier driving. Despite many industrial innovations and academic progresses have been made, we find a lack

2、 of fully context-aware smart cars. This study presents a general architecture of smart cars from the viewpoint of context- awareness. A hierarchical context model is proposed for description of the complex driving environment. A smart car prototype including software platform and hardware infrastru

3、ctures is built to provide the running environment for the context model and applications. Two performance metrics were evaluated: accuracy of the context situation recognition and efficiency of the smart car. The whole response time of context situation recognition is nearly 1.4 s for one person, w

4、hich is acceptable for non-time critical applications in a smart car.Key words: Smart car, Intelligent vehicle, Context-aware, Ubiquitous computing.doi:10.1631 /jzus.A0820154 Document code: A CLC number: TP39INTRODUCTION Cars are becoming important private places frequently used in daily life. Howev

5、er, they also bring many problems, such as traffic congestions and accidents. A smart car aims at assisting its driver with easier driving, less workload and less chance of getting injured (Moite, 1992). For this purpose, a smart car must be able to sense, analyze, predict and react to the road envi

6、ronment, which is the key feature of smart cars: context-awareness. Lots of technologies have been developed in the past decade, such as intelligent transportation systems(ITS) (Wang et al., 2006) and the advanced driver assistant system (ADAS) (Kkay and Bergholz,2004). However, current smart cars a

7、re not really context-aware. Only a few types of the information of road environments, which is called contexts, are utilized. Besides, most of current smart cars lack complex reasoning. These drawbacks limit the smart cars ability of assisting the driving task efficiently and safely. This research

8、focuses on how to build a context-aware smart car. The remainder of this paper is organized as follows. Section 2 introduces the related work on smart cars. A general description of a smart car is given in Section 3. Section 4 proposes a hierarchical context model for comprehensive definition and cl

9、assification of information in a smart car environment. The smart car prototype, including the hardware infrastructure and software platform, is presented in Section 5. The performance evaluation is shown in Section 6 and the conclusions are given in Section 7.RELATED WORK In the past decade, many r

10、esearches from academic and industrial communities have been made on smart cars. The following is a summary of the major progresses in this field. (1) New manufacturing technology. MIT Media Lab presents a conceptual car, City Car (MIT, 2004), a lightweight electric vehicle. This car employs fully i

11、ntegrated in-wheel electric motors and suspension systems, which are self-contained, digitally controlled, and reconfigurable. With the wireless connectivity and a Google-like information grid, drivers could use the information to navigate in a very intelligent way. (2) Driver assistant system. Auto

12、motive manu-factures implement many novel ideas in their newest series of concept cars. BMWs ConnectedDrive includes BMW Assist, BMW Online and driver assistance systems, supporting lane change warning and parking assistant (Hoch et al., 2007). Mercedes-Benz is developing an intelligent driver assis

13、tance system that utilizes stereo cameras and radar sensors to monitor the surroundings around the car (Benz, 2007).Volvos CoDriver is an intelligent assistant that co-ordinates information, studies the traffic situation and assists the driver (Volvo, 2007). Lexus provides advanced active safety tec

14、hnologies on its LS-series,including an advanced pre-collision system, dynamic driving, electronic brake assistance, and park-assistance systems (Lexus, 2007). (3) Collision avoidance system. The SAVE-IT project develops a central component that monitors the roadway, the states of the vehicle and th

15、e driver, with evaluation of the potential safety benefits (Lee etal., 2004). The Cybercars project addresses navigation, obstacle avoidance and platooning (Parent and Fortelle, 2005). The SAFESPOT project aims at expanding the time horizon for acquiring safety relevant information and improving pre

16、cision, reliability and quality of driving (Giulio, 2007). The prevent project develops preventive safety technologies and in-vehicle systems, which sense the potential danger and take the drivers state into account (Matthias,2006). (4) Driver-vehicle interface. The Adaptive Integrated Driver-vehicl

17、e Interface (AIDE) project tries to maximize the efficiency and safety of advanced driver assistance systems, while minimizing the workload and distraction imposed by in-vehicle information systems (Kutila et al., 2007). The Communication Multimedia Unit Inside Car (COMUNICAR) project aims at design

18、ing aneasy-to-use on-vehicle multimedia human-machine interface. An information manager collects the feed back information and estimates the drivers workload according to the current driving and environment situation (Bellotti et al., 2005). (5) Driver behavior recognition. The driver plays an impor

19、tant role in a smart car. Machine learning and dynamical graphical models, such as HMM (Oliver and Pentland, 2000), Gaussian Mixture Modeling (GMM) (Miyajima et al., 2007) and the Bayesian network (Kumagai and Akamatsu, 2006), can be applied for modeling and recognizing driver behaviors. (6) Communi

20、cation and cooperation. The Car-TALK project enables information transmitting among cars in the vicinity (Reichardt et al., 2002).The COM2REACT Project (2006) establishes a cooperative and multi-level transport virtual sub-center by vehicle-to-vehicle communication and vehicle-to-centre communicatio

21、n. The COOPERS Project (2006) provides local situation information, traffic and infrastructure status information via a dedicated infra-structure to support vehicle communication link. The COVER Project (2006) develops semantic-driven cooperative systems with the main focus on communication between

22、the infrastructure and vehicles. The I-WAY Project (Rusconi et al., 2007) designs an intelligent cooperative system, which provides real-time information from other vehicles in the vicinity and roadside equipments to improve drivers responses.The WATCH-OVER Project (2006) develops a cooperative syst

23、em for the prevention of road accidents involving vulnerable road users, such as motorcycles,bicycles, and pedestrians. The Cooperative Vehicle-Infrastructure Systems (CVIS) Project (2006) creates a unified technical solution allowing all vehicles and infrastructure elements to communicate with each

24、 other in a continuous and transparent way. 7) Safety in vehicles. The Secure Vehicular Communication (SEVECOM) Project provides a full definition and implementation of security requirements for vehicular and inter-vehicular communications (Panos et al., 2006). Vehicular Ad-hoc Network (VANET) secur

25、ity also partly addresses the safety in vehicles (Magda et al., 2002; Hubaux et al., 2004; Raya and Hubaux, 2005; Bryan and Adrian, 2005),which gives the problem statement and proposes the outline of a general solution for VANET. However, we found that most of the work listed above is not fully cont

26、ext-aware. Current work usually focuses on special practical technologies, such as communication, sensing and driver assistance. In addition, the reasoning of contexts for further analysis is not put enough emphasis on. These will limit smart cars to be cars with certain accessories, so different re

27、quirements will result in different smart cars. There is a lack of a common consensus and comprehensive understanding of smart cars in a holistic view. This study attempts to build a smart car from the viewpoint of context-awareness in a bottom-up manner. We want to build a general theoretical found

28、ation and an infrastructure framework for a smart car. All the contexts that can characterize an entity of the driving environment will be collected and defined. Reasoning will play an important role in complex situation analysis. In such a smart car, we can develop different services and applicatio

29、ns without much modification to the current architecture.GENERAL ARCHITECTURE A smart car is a comprehensive integration of many different sensors, control modules, actuators,and so on (Wang, 2006). A smart car can monitor the driving environment, assess the possible risks, and take appropriate acti

30、ons to avoid or reduce the risk. A general architecture of a smart car is shown in Fig.1. (1)Traffic monitoring. A variety of scanning technologies can be used to recognize the distance between the car and other road users. Active environments sensing in- and out-car will be a general capability in

31、the near future (Tang et al., 2006). Lidar-radar- or vision-based approaches can be used to provide the positioning information. The radar and lidar sensors provide information about the relative position and relative velocity of an object. Multiple cameras are able to eliminate blind spots, recogni

32、ze obstacles, and record the surroundings. Besides the sensing technology described above, the car can get traffic information from the Internet or nearby cars. (2)Driver monitoring. Drivers represent the highest safety risk. Almost 95% of the accidents are due to human factors and in almost three-q

33、uarters of the cases human behaviour is solely to blame (Rau,1998). Smart cars present promising potentials to assist drivers in improving their situational awareness and reducing errors. With cameras monitoring the drivers gaze and activity, smart cars attempt to keep the drivers attention on the r

34、oad ahead. Physiological sensors can detect whether the driver is in good condition. (3)Car monitoring. The dynamics of a car can be read from the engine, the throttle and the brake. These data will be transferred by controller area networks (CAN) to analyze whether the car functions normally. (4)As

35、sessment module. It determines the risk of the driving task according to the situation of the traffic, driver and car. Different levels of risks will lead to different responses, including notifying the driver through the Human Machine Interface (HMI) and taking emergency actions by car actuators. (

36、5) HMI. It warns the driver of the potential risks in non-emergent situations. For example, a tired driver would be awakened by an acoustic alarm or vibrating seat. Visual indications should be applied in a cautious way, since a complex graph or a long text sentence will seriously impair the drivers

37、 attention and possibly cause harm. (6)Actuators. The actuators will execute specified control on the car without the drivers commands. The smart car will adopt active measures such as stopping the car in case that the driver is unable to act properly, or applying passive protection to reduce possib

38、le harm in abrupt accidents, for example, popping up airbags.HIERARCHICAL CONTEXT MODEL Contexts are information collected when monitoring the roadway, the car and the driver. To implement a context-aware smart car, we must begin with the context analysis. We develop a hierarchical context model, wh

39、ich is the basis of representation and analysis of the smart car environment.Hierarchical context model We categorize context data into three layers according to the degree of abstraction and semantics:the sensor layer, the context atom layer and the context situation layer, as shown in Fig.2. The s

40、ensor layer is the source of context data, the context atom layer serves as an abstraction between the physical world and semantic world, and the context situation layer provides description of complex facts with fusion of context atoms.Context atoms: ontology definition Each sensor corresponds to a

41、 type of context atom. For each type of context atom, a descriptive name must be assigned for applications to use the contexts. We use ontology to define the name to guarantee the semantic understanding and sharing in smart cars. We use three ontologies as shown in Fig.3(1) Ontology for environment

42、contexts. The environmental contexts are related to physical environments. The ontology includes the description of weather, road surface conditions, traffic information, road signs, signal lamps, network status, etc.(2) Ontology for car contexts. The car ontology includes three parts: the power sys

43、tem, the security system and the comfort system. The power system concerns the engine status, the accelerograph, the power (gasoline), etc. The security system includes factors related to safety of the car and the driver, such as the status of the air bag, the safe belt, the anti-lock braking system

44、 (ABS), the reverse-aids, the navigation system, and the electronic lock. The comfort system is about entertainment devices, the air conditioner, windows, etc.(3) Ontology for driver contexts. The driver contexts are about the drivers physiological conditions, including the heart beat, blood pressur

45、e, density of carbon dioxide, diameter of pupils, etc. The information is used to evaluate the health and mental statuses of the driver for determining whether he/she is able to continue driving. To use the context atoms, the subscription and publication mechanisms are employed. Those applications i

46、nterested in specific contexts atoms will be added to the subscriber list, along with information on how to publish context to them. Once a subscribed context changes, the new data will be delivered to the subscribing application.Context situation: training and recognition Context situation recognit

47、ion is a reasoning process and should be real time. This research uses a pattern-based inference engine and includes two parts:offline statistic-based situation pattern training and online situation recognition (Fig.5). The training phase is used to learn the statistical relationship between context

48、 atoms and situations and hence to generate the pattern of every single situation. The online recognition phase is used to recognize the current situation according to its pattern in the running time of a smart car. 附錄B 具有情景感知的智能汽車：從模型到原型的發(fā)展摘要：由于智能汽車到處都應(yīng)用著微機(jī)，所以這是有前途的領(lǐng)域。在敏感環(huán)境中主要就是為了智能汽車更安全和更容易的駕駛。盡管許

49、多工業(yè)創(chuàng)新和學(xué)術(shù)研究上取得了很大的進(jìn)展，但是我們發(fā)現(xiàn)充分缺乏具有情景感知的智能汽車。本研究闡訴的總體結(jié)構(gòu)是智能汽車的語境方面。其中一方面描述復(fù)雜的駕駛環(huán)境的模型。智能汽車原型的內(nèi)置設(shè)施包括具有情景感知的軟件模型和提供應(yīng)用程序運(yùn)行環(huán)境的硬件。對其進(jìn)行評估有兩個(gè)性能指標(biāo)：對語境、情景識別精度和效率。對整個(gè)語境識別所的響應(yīng)時(shí)間大約是一個(gè)人的1.4陪，在非時(shí)間關(guān)鍵型智能車的應(yīng)用程序中是可接受的。關(guān)鍵詞:智能汽車、智能車輛、環(huán)境敏感、無處不在的微機(jī)數(shù)字對象唯一標(biāo)識符：10.1631/浙江大學(xué)科學(xué)雜志。A0820154文檔代碼:TP39 CLC.介紹 在日常生活中汽車將成為私人經(jīng)常使用中重要的部分。然而，

50、他們也帶來很多問題，如交通擁擠和事故。智能汽車的目的是協(xié)助駕駛員更容易駕駛，減少駕駛員的工作量和受傷的機(jī)會。為了這個(gè)目的，一個(gè)智能的汽車必須能夠感知、分析、預(yù)測和反應(yīng)道路的環(huán)境，智能汽車的關(guān)鍵特征是語境意識。 在過去的十年中已經(jīng)應(yīng)用許多技術(shù)，如智能交通運(yùn)輸系統(tǒng)和先進(jìn)的駕駛輔助系統(tǒng)。然而，目前的智能汽車是不能真正感知情景。只利用在少數(shù)道路的環(huán)境類型，這被稱為背景。此外，目前的智能汽車缺乏復(fù)雜的推理。這些缺點(diǎn)限制了輔助駕駛?cè)蝿?wù)的智能車的能力和安全。本文研究的重點(diǎn)是如何研制出具有情景感知的智能汽車。 本文的一下部分安排如下：第二2節(jié)介紹了智能車的相關(guān)工作。在第3節(jié)對智能小車進(jìn)行描述。第4節(jié)介紹綜合應(yīng)

51、用在智能車運(yùn)行環(huán)境中具有情景感知和分析信息的模型。在第5節(jié)介紹智能汽車的原型，包括硬件設(shè)施和軟件平臺。在第6節(jié)和7節(jié)中給出績效評估的結(jié)論。相關(guān)工作 在過去的十年中，許多學(xué)術(shù)界和產(chǎn)業(yè)界已經(jīng)在研究智能汽車。以下是這一領(lǐng)域的主要進(jìn)展的綜述。（1）新的制造技術(shù)。麻省理工學(xué)院媒體實(shí)驗(yàn)室研制出了一個(gè)概念車，城市車（麻省理工學(xué)院，2004），一個(gè)輕量級的電動(dòng)車輛。這輛車采用了完全集成在車輪的電動(dòng)馬達(dá)和懸掛系統(tǒng)，都是獨(dú)立的、數(shù)字控制的、可重構(gòu)的。附帶無線連接和一個(gè)谷歌信息網(wǎng)格，司機(jī)可利用這些明確路況。（2） 汽車馬努建筑中實(shí)現(xiàn)許多新穎的想法在他們的最新系列的概念車。寶馬的互聯(lián)駕駛包括寶馬協(xié)助、寶馬在線和司機(jī)輔

52、助系統(tǒng),支持換車道警告和停車助理(Hoch et al.,2007)。梅賽德斯-奔馳是一個(gè)智能司機(jī)援助系統(tǒng),利用立體攝像機(jī)和雷達(dá)傳感器來監(jiān)視四周車(奔馳,2007)。沃爾沃的副駕駛員是個(gè)聰明的助理,負(fù)責(zé)協(xié)調(diào)信息,研究交通狀況,協(xié)助司機(jī)(沃爾沃,2007)。雷克薩斯對其LS系列提供了先進(jìn)的主動(dòng)的安全技術(shù),其中包括一個(gè)先進(jìn)的預(yù)碰撞系統(tǒng),動(dòng)態(tài)驅(qū)動(dòng)、電子剎車援助,公園援助系統(tǒng)(雷克薩斯,2007)。（3） 防撞系統(tǒng)。塞維特項(xiàng)目開發(fā)的一個(gè)重要組成部分是監(jiān)視道路，對車輛和駕駛員的狀態(tài)和潛在的安全效益評價(jià)（Lee等人，2004）。智能交通車輛項(xiàng)目地址導(dǎo)航，避障和排隊(duì)（Parent和Fortelle，2005

53、）。這個(gè)項(xiàng)目主要是在擴(kuò)展時(shí)間范圍，獲取安全相關(guān)的信息,提高測量精度、可靠性和質(zhì)量的驅(qū)動(dòng)(朱里奧,2007)。（4） 車輛接口。車輛接口(助手)項(xiàng)目目的是最大化效率和使高級駕駛輔助系統(tǒng)更為安全,同時(shí)最小化工作負(fù)載和干擾車載信息系統(tǒng)實(shí)施的(Kutila等人,2007)。通信多媒體單元內(nèi)的車(通信)項(xiàng)目旨在設(shè)計(jì)懲罰使用機(jī)上的多媒體人機(jī)界面。一個(gè)信息經(jīng)理收集反饋信息,根據(jù)當(dāng)前的駕駛和環(huán)境情況估計(jì)司機(jī)的工作量(Bellotti 等人,2005)。（5） 司機(jī)的行為識別。在一個(gè)智能汽車上，驅(qū)動(dòng)程序扮演重要的角色。機(jī)器實(shí)驗(yàn)和動(dòng)態(tài)圖形模型,比如隱馬爾可夫模型(Oliver 和 Pentland,2000),高

54、斯混合模型(Miyajima等人,2007年)和貝葉斯網(wǎng)絡(luò)(Kumagai和Akamatsu,2006),可廣泛應(yīng)用于建模和識別司機(jī)的行為。（6）溝通和合作。Car-TALK用于傳輸信息在汽車附近(Reichardt等人,2002)。COM2REACT(2006)建立了一個(gè)合作的、多層次的運(yùn)輸車輛,車輛虛擬分中心通過溝通和車輛中心通信。COOPERS項(xiàng)目(2006)提供了當(dāng)?shù)厍闆r信息,通過專門結(jié)構(gòu)支持把交通、基礎(chǔ)設(shè)施狀態(tài)信息和車輛通信鏈接。封面項(xiàng)目(2006)開發(fā)語義驅(qū)動(dòng)的合作系統(tǒng)與主要集中在基礎(chǔ)設(shè)施和車輛之間的通信。 COVER 項(xiàng)目(Rusconi等人,2007)設(shè)計(jì)了一種智能合作系統(tǒng),它

55、從其他車輛在附近和路邊設(shè)備提供了實(shí)時(shí)信息來提高司機(jī)的反應(yīng)。I-WAY項(xiàng)目(2006)發(fā)展合作系統(tǒng)防止交通事故涉及弱勢道路使用者,如摩托車、自行車和行人。合作車輛基礎(chǔ)設(shè)施系統(tǒng)(減振系統(tǒng))項(xiàng)目(2006)創(chuàng)建了一個(gè)統(tǒng)一的技術(shù)解決方案允許所有車輛和基礎(chǔ)設(shè)施元素相互溝通在一個(gè)連續(xù)的和透明的方式。7）車輛安全。安全的車載通信項(xiàng)目提供了一個(gè)完整的定義和實(shí)施車輛安全要求，和其他車輛間通信（Panos等人，2006）。車載Ad Hoc網(wǎng)絡(luò)的安全也部分地解決了安全車（Magda等人, 2002; Hubaux等人, 2004; Raya和Hubaux, 2005; Bryan 和Adrian, 2005），根據(jù)

56、VANET給出了問題的陳述并提出了解決的方案。然而，我們發(fā)現(xiàn)大部分的工作是并沒有充分語境感知。目前的工作通常集中在特殊的實(shí)用技術(shù)，如通信，傳感和驅(qū)動(dòng)程序的幫助。此外，為進(jìn)一步分析語境的推理是不足夠被重視的。這將限制智能汽車得某些配件，所以要求不同，會導(dǎo)致不同的智能汽車。對智能車的綜合理解有一個(gè)缺乏共識和全面的觀點(diǎn)。本研究試圖從一個(gè)自底向上的方式，上下文感知的角度建立一個(gè)智能車。我們想建立一個(gè)一般性的理論基礎(chǔ)和智能汽車的基礎(chǔ)框架。所有可以表征的駕駛環(huán)境的實(shí)體將被收集和定義的上下文。對復(fù)雜的形勢的分析、推理將發(fā)揮重要的作用。在這樣一個(gè)智能車，我們可以擁有不同的服務(wù)和應(yīng)用程序。整體架構(gòu) 智能汽車是許

57、多不同的傳感器、控制模塊和執(zhí)行器等的綜合集成，（王，2006）。智能車的駕駛環(huán)境監(jiān)測，評估可能的風(fēng)險(xiǎn)，并采取適當(dāng)?shù)男袆?dòng)，以避免或減少風(fēng)險(xiǎn)。（1） 交通監(jiān)控。各種掃描技術(shù)可用于識別距離車和其他道路使用者之間。積極的環(huán)境感知和汽車將在不久的將來的一般能力（唐等人。，2006）。激光雷達(dá)和視覺為基礎(chǔ)的方法可以用來提供定位信息。雷達(dá)和激光雷達(dá)傳感器提供的相對位置和相對速度信息的對象。多攝像機(jī)能夠消除盲點(diǎn)，認(rèn)識障礙，并記錄環(huán)境。除了上述的傳感技術(shù)，汽車可以得到交通信息從互聯(lián)網(wǎng)上或附近的汽車。（2） 驅(qū)動(dòng)程序監(jiān)控。司機(jī)代表最高的安全風(fēng)險(xiǎn)。幾乎95%的事故是由于人的因素和幾乎四分之三的人的行為是完全怪（RA

58、U，1998）。智能汽車的有前途的潛力，協(xié)助駕駛員提高態(tài)勢感知和減少錯(cuò)誤。攝像機(jī)監(jiān)控駕駛員的視線和活動(dòng)，智能汽車試圖讓駕駛員的注意力在前進(jìn)的道路上。生理傳感器可以檢測司機(jī)是否處于良好狀態(tài)。（3） 車輛監(jiān)控。一輛汽車的動(dòng)力學(xué)可以從機(jī)讀，油門和剎車裝置。這些數(shù)據(jù)將通過控制器區(qū)域網(wǎng)絡(luò)（CAN）轉(zhuǎn)移分析汽車是否功能正常。（4） 評估模塊。它決定根據(jù)交通狀況的駕駛?cè)蝿?wù)的危險(xiǎn)，司機(jī)和汽車。不同級別的風(fēng)險(xiǎn)會產(chǎn)生不同的反應(yīng)，包括通知司機(jī)通過人機(jī)界面（HMI）和汽車執(zhí)行器采取緊急行動(dòng)。（5）人機(jī)界面。它警告說，在非緊急情況下的潛在風(fēng)險(xiǎn)的驅(qū)動(dòng)程序。例如，一個(gè)累了司機(jī)會吵醒的報(bào)警聲或振動(dòng)座椅。視覺指示應(yīng)以謹(jǐn)慎的方式

59、，從一個(gè)復(fù)雜的圖形或長文本的句子會嚴(yán)重?fù)p害駕駛員的注意力和可能造成的危害。（6） 執(zhí)行器。執(zhí)行器將執(zhí)行指定的控制對汽車沒有司機(jī)的命令。智能汽車將采取積極措施，如停止的情況下，汽車的驅(qū)動(dòng)程序無法正確動(dòng)作，或采用被動(dòng)保護(hù)減少突發(fā)事故可能造成的危害，例如，彈出安全氣囊。語境感知的模型 環(huán)境監(jiān)測收集汽車、司機(jī)和路面信息時(shí)，為了實(shí)現(xiàn)情景感知的智能車，我們必須從語境分析出發(fā)。我們開發(fā)了一個(gè)層次的情景感知的模型，這是表示和分析智能車的環(huán)境的基礎(chǔ)上。語境感知的模型 根據(jù)抽象程度和語義，我們將語境感知分為三層：傳感器層，背景層和背景層原子的情況下，傳感器層的感知數(shù)據(jù)源，感知原子層作為物理世界和意義世界之間的抽象，和感知的情況下提供復(fù)雜的事實(shí)描述