24 2 2
i
ii 2 1. 2. 15 87% 2 1. 2.
An Analysis of Individual Characteristics of Driving Behavior Using Accelerometers Abstract iii Akimasa NAKASHIMA Traffic system is composed by interaction between many cars and very complicated system. It is difficult to evaluate efficiency of the system. For such complicated system, Multi-Agent Traffic Simulator is used to evaluate. But traffic simulator based on Multi-Agent simulator needs appropriate agent model. Driving agent is designed by analysis of driver in real world and modeling it. So far, the main way of analyzing driving behavior is interview to drivers and using sensor in car. But these method can t measure movement of drivers and identify the difference of cars which behave same action. In previous research, a paper suggests that by Accelerometers identify that difference. In paper, the author measures steering behaviors by acceleration of the wrists. In short, we can measure behaviors if driver can t remember it. And this way is possible to measure behavior which is not measurable by sensor in car.,such as disengaging a hand from handle. However, the paper focused on detecting steering behavior different from that of good drivers, and the proposed method is not used for analysis of drivers behavior. Therefore, this research focuses on analysis on individuality of drivers behavior using accelerometers. Here individuality means behavioral difference among individual drivers. The analysis targets especially steering behaviors in turning right. In analysis of individuality, there are 2 problems as follows. 1. Quantitartive index that indicates difference steering behavior Considering individuality, it is required to show quantitative difference in drivers steering difference. However, it is difficult to explain it by quantitative index though existence of difference in steering behavior can be understood intuitively. 2. Method for analysis of individuality Researchers have not considered much about individuality in driveling behavior has not been considered before. Therefore, firstly, it is necessary to
iv show existence of individuality in driving behavior. Considering the problems above, I analyzed individuality in driving behavior, using accelerometers. First, I had an driving experiment using a driving simulator with 15 examinees. Then watching the videos that recorded examinees driving behavior, I extracted patterns of steering behavior in turning right. Next, I extracted feature quantity in accelerometers that enables categorization into the extracted patterns. Comparing waveforms of each pattern, I decided to use standard deviation in accelerometers in both hands. Finally, I created the classifier and classified the recorded data. The classifier could classify the data correctly with classification accuracy of about 87%. Then, as an index in steering behavior, I compared dispersion in acceleration of a driver s both hands using the mentioned feature quantity. Contributions of this research are following 2 things. 1. Proposal of the feature quantity of acceleration that explains difference in steering patterns Classifying steering patterns visually from drivers steering behaviors in turning right, I determined several patterns of steering behavior. I extracted the feature quantity that can be classified significantly by comparison of the values of acceleration in the steering patterns. 2. Analysis by comparison of drivers using acceleration Using the feature quantity in accelerometers, I compared the feature quantity in steering behavior in turning right of each driver. The comparison showed individuality because each driver had different dispersion in the feature quantity.
1 1 2 & 2 2.1............................. 2 2.2............................... 3 3 4 3.1.................... 4 3.2............................... 6 4 7 4.1............................ 7 4.2............................ 10 5 12 5.1....................................... 12 5.2............................... 13 5.2.1........................... 13 5.2.2................... 20 5.3.......................................... 22 6 23 23 24
1 [1, 2] 2 3 1
4 5 6 2 & 2.1 [3, 4] SOUND [5] SOUND Q-K - MATSim[1] MATSim 2
Hallé HESTIA [2] [6] tiss-net [7] 2.2 Pipes [8] v(t) = A (v front (t 1) v(t)) v(t) t v front (t) t A Nagel [9] v leftfront v(t) or v rightfront(t) v(t) v leftfront (t) v rightfront (t) t 3
[10] 3 3.1 GPS [11] [12] Cheng [13] Albinali 4
1: [14] HASC Challenge [15] [16] [17, 18] 1 SVM SVM 5
図 2: 加速度センサの装着箇所 (右手) する 得られた加速度データに対して 一定周期における値を切り出す時間窓 を適用し データの切り分けを行う 次に 時間窓内の加速度の平均 軸と利 き手の左右に関する相関 周波数密度の 3 点における特徴量の抽出を行う 得 られた特徴量を教師データとして 1 クラス SVM に学習させる そして 学習 して得られた識別器を利用し 逸脱動作を検出する 一般運転者に対して 同 様に装着したセンサから加速度データを計測し 特徴量を抽出する 抽出した 特徴量を SVM に入力し SVM が入力された特徴量が外れ値であるかを判断す ることで 逸脱動作の検出を試みている 3.2 ハンドル操作の計測 本論文では 運転操作における人間の挙動を分析するが 具体的には 加速 度センサによる計測との親和性を考慮して 人間のハンドル操作に注目する1) 従って 人間のハンドル操作の計測方法が問題となる この問題に対しては 従 来 車両のハンドルなどの操作系デバイスに各種センサを取り付け 人間の操 作挙動を計測する試みが多い [19] しかし 計測専用車両の準備にかかるコスト が大きく また操作系デバイスに入力されるデータのみの利用に限られる 本 論文では 操作系デバイスの状態を変化させる人間の操作挙動に注目するため 1) ブレーキやアクセルペダルの踏み込みの動きは緩やかなケースが多いため 実際に計測した ところ 加速度の波形中に視認可能な変化を見出す事が困難であった 6
3.1 [17] ATR-promotions WAA-006 1) cm 1cm x y z 3 [18] 2 4 4.1 5 1) http://www.atr-p.com/sensor06.html 7
3: 3 3 4 4 5 8
図 4: 右手だけ持ち替えを行うパターン 図 5: 両手で持ち替えを行うパターン 9
6: 5 4.2 4.1 6 7 x y z 10
7: 6 8 11
8: 5 4 5.1 12
9: 9 15 3 18 5.2 5.2.1 5.1 18 15 270 10 11 12 4.1 13
10: 10 11 12 14
11: 12: 15
13: 4.1 4.2 13 16
14 99msec 33 51msec 17 13 3000msec 11 15 13 (1000,1000) 17
14: 18
15: 19
1: SVM no right both no 46.25 0.5 3.75 right 12.75 9 17.25 both 1.75 6.75 36 15 3 SVM 5.2.2 270 105 55 108 1) 2 268 2 ( ) SVM 4 3 68% 23% SVM 16 no right both no both right no both right 1) 20
16: 2: no right both no 50.75 0.5 2.75 right 6.75 16.5 5.25 both 2.25 0 49.25 2 x,y,z 2 2 x,y,z 21
17: 87% 28.5 16.5 58% 5.3 5.2 17 5 6 6 5 22
17 6 1. 2. 23
[1] Balmer, M., Meister, K., Rieser, M., Nagel, K. and Axhausen, K. W.: Agent-based Simulation of Travel Demand: Structure and Computational Performance of MATSim-T, Proc. of the 2nd TRB Conference on Innovations in Travel Modeling, pp. 1 30 (2008). [2] Hallé, S. and Chaib-draa, B.: A Collaborative Driving System based on Multiagent Modelling and Simulations, Transportation Research Part C, Vol. 13, pp. 320 345 (2005). [3], :,, Vol. 12, pp. 731 738 (1995). [4] :, (2006). [5] Yoshii, T. and Kuwahara, M.: SOUND: A Traffic Simulation Model for Oversaturated Traffic Flow on Urban Expressways, Proc. of the 7th World COnference on Transporation Research (1995). [6], : INSPECTOR,, Vol. 39, No. 4, pp. 72 83 (2004). [7],, :,, Vol. 25, No. 3, pp. 815 821 (2008). [8] Pipes, L. A.: An Operational Analysis of Traffic Dynamics, Journal of Applied Physics, Vol. 24, pp. 274 281 (1953). [9] Nagel, K., Wolf, D. E., Wagner, P. and Simon, P.: Two-lane traffic rules for cellular automata: A systematic approach, Phys. Rev. E, Vol. 58, pp. 1425 1437 (1998). [10] Hattori, H., Nakajima, Y. and Ishida, T.: Learning From Humans: Agent Modeling With Individual Human Behaviors, Systems, Man and Cybernetics, Part A: Systems and Humans, IEEE Transactions on, Vol. 41, No. 1, pp. 1 9 (2011). [11] : ( : ),, Vol. 23, No. 5, pp. 611 616 24
(2008). [12] : ( ), -, Vol. 28, No. 2, pp. 43 54 (2011). [13] Chang, K.-H., Chen, M. Y. and Canny, J.: Tracking free-weight exercises, Proceedings of the 9th international conference on Ubiquitous computing, pp. 19 37 (2007). [14] Albinali, F., Intille, S., Haskell, W. and Rosenberger, M.: Using wearable activity type detection to improve physical activity energy expenditure estimation, Proceedings of the 12th ACM international conference on Ubiquitous computing, pp. 311 320 (2010). [15] :,, Vol. 52, No. 1, pp. 32 36 (2011). [16] Murao, K. and Terada, T.: A motion recognition method by constancydecision, Wearable Computers (ISWC), 2010 International Symposium on, pp. 1 4 (2010). [17],,,,,, : ( ),. D,, Vol. 91, No. 4, pp. 1115 1129 (2008). [18],,,,, :,, Vol. 23, No. 3, pp. 105 116 (2008). [19],,,,, :,, Vol. J90-D, No. 4, pp. 1115 1123 (2007). [20] Kawaguchi, N., Ogawa, N., Iwasaki, Y., Kaji, K., Terada, T., Murao, K., Inoue, S., Kawahara, Y., Sumi, Y. and Nishio, N.: HASC Challenge: gathering large scale human activity corpus for the real-world activity understandings, Proceedings of the 2nd Augmented Human International Conference, AH 11, New York, NY, USA, ACM, pp. 27:1 27:5 (2011). [21] Guyot, P., Drogoul, A. and Honiden, S.: Power and negotiation: lessons 25
from agent-based participatory simulations, Proceedings of the fifth international joint conference on Autonomous agents and multiagent systems, pp. 27 33 (2006). [22] Sempé, F., Nguyen-Duc, M., Boucher, A. and Drogoul, A.: An artificial maieutic approach for eliciting experts knowledge in multi-agent simulations, Proceedings of the fourth international joint conference on Autonomous agents and multiagent systems, pp. 1361 1362 (2005). 26