ITS [1] [4] [5] DTW Dynamic Time Warping [6] 1 1,000 [7] CCTV Closed-Circuit Television [8] [10] CCTV CCTV Forren Chen c 016 Inform

Size: px

Start display at page:

Download "ITS [1] [4] [5] DTW Dynamic Time Warping [6] 1 1,000 [7] CCTV Closed-Circuit Television [8] [10] CCTV CCTV Forren Chen c 016 Inform"

みいかうづき
5 years ago
Views:

1 Dynamic Time Warping 1,a) ITS: Intelligent Transportation Systems 1 DTW Dynamic Time Warping 0.9 DTW Dynamic Time Warping Design and Initial Evaluation of Acoustic Vehicle Count System utilizing Dynamic Time Warping Shigemi Ishida 1,a) Song Liu 1 Kohei Mimura 1 Shigeaki Tagashira Akira Fukuda 1 Abstract: Vehicle counting is one of the fundamental tasks in the intelligent transportation systems (ITS). Although automatic vehicle counters have been proposed to retrieve realtime traffic data, current automatic vehicle counters are suffered from high deployment costs, resulting in limited number of deployments. In this paper, we present a vehicle counter using sidewalk microphones that are easily deployed at a low cost. Our vehicle counter only relies on two microphones and counts vehicles using sound map, which is a timedifference map of vehicle sound on the two microphones. We developed a vehicle count algorithm using sound map. Our vehicle count algorithm performs template matching utilizing dynamic time warping (DTW) to adapt to vehicle speed variation. We herein describe the design and implementation of our vehicle count system. Experimental evaluations reveal that our vehicle count system successfully counted vehicles with a precision of 0.9. Keywords: Vehicle count, DTW (dynamic time warping), acoustic sensing, sound map ISEE, Kyushu University, Fukuoka , Japan Faculty of Informatics, Kansai University, Osaka, , Japan a) ishida@f.ait.kyushu-u.ac.jp ITS: Intelligent Transportation Systems ITS ETC Electronic Toll Collection System ITS ITS c 016 Information Processing Society of Japan 1

2 ITS [1] [4] [5] DTW Dynamic Time Warping [6] 1 1,000 [7] CCTV Closed-Circuit Television [8] [10] CCTV CCTV Forren Chen c 016 Information Processing Society of Japan

3 M 1 M Sound Retriever LPF LPF Soundmap Sound Mapper Vehicle Counter O d 1 d D/ D/ M 1 M L x Fig. 1 1 Overview of acoustic vehicle count system Fig. Microphone setup [1] [3] Barbagli [4] [11], [1] Sound Retriever Sound Mapper Vehicle Counter 3 LPF Low-pass Filter LPF DTW Dynamic Time Warping 3. M 1 M D L x = ± t max c ± t max = ± D (1) c D D L D L LPF.0 khz [13], [14] LPF.5 khz c 016 Information Processing Society of Japan 3

4 3.3 M 1 M D L x d 1 d ( d 1 = x + D ), () ( d = x D ) (3) t c t = d 1 d c = 1 ( x + D ) c ( x D ) (4) (4) s 1 (t) s (t) R(t) R(t) = s 1 (τ) s (t + τ) dτ. (5) t s 1 (t) = s (t + t) R(t) t = t R(t) GCC: Generalized Cross- Correlation [15] GCC 3 Sound Delay t S S 3.4 DTW Dynamic Time Warping Sound delay t [ms] Fig. 3 Time t [s] Example of sound map DTW S DTW S S S S 4 DTW S (1) DTW DTW (1) () (1)+() (3) DTW S 5 S t = ± t max L t det L L t t det [k] L t det = ( t det [k]). (6) k c 016 Information Processing Society of Japan 4

5 情報処理学会研究報告 Sound Map Sound Map (1) () (3) (4)... Template (Left to Right) match with (1) with (1)+() with (3)... Partially match Template (Right to Left) match with (1) with (1)+() with (1)+()+(3) with (4)... Template 図 5 DTW を用いたテンプレートマッチングにおける誤検出図 4 DTW を用いた車両カウントの概要 Fig. 5 False positive detection on template Fig. 4 Vehicle count using DTW matching using DTW -lane Road Recorder Microphones 図 7 図 6 評価環境車両検出の例右から左へ進む車両 Fig. 7 Example of vehicle detections (right to left vehicles) Fig. 6 Experiment setup した図 5 に示すように誤検出部のサウンドマップには到達時マイクロフォンの間隔 D は予備実験の結果から 50 cm 間差 t が tmax から tmax まで変化する遷移部分がとした予備実験ではマイクロフォンの間隔を存在しないため L ノルムが大きくなる車両検出部のサ 150 cm と変化させながら約 0 分間に渡って走行音データウンドマップ系列長は一定ではないため閾値との比較のを収集し検出率を比較した前に系列長で L ノルムを正規化する 4. 評価録音した走行音を元にサウンドマップを描き提案する車両カウント手法を適用して車両を検出した図 7 に提案システムによる車両検出の例を示す赤い縦線が車両の提案システムの有効性を検証するための初期的評価とし通過を示している提案システムでは車両の進行方向別にて九州大学伊都キャンパス内の道路において実証評価をテンプレートマッチングを行って車両を検出するためそ行ったれぞれの方向の検出結果を加算して総合の検出結果を算出した 4.1 評価環境検出結果をカメラ画像と比較することで True Positive 図 6 に評価環境を示す対象となる道路は片側 1 車線合 TP False Negative FN False Positive FP True 計車線の道路である台のマイクロフォンを高さ 1 m Negative TN の回数を評価した TP FN FP TN は道路の中心から約 m 離れた歩道上に設置し IC レコーそれぞれ走行車両があるときに車両をカウントした場合ダで車両の走行音を約 5 分間録音した IC レコーダは走行車両があるときに車両をカウントしなかった場合走 SONY 社製 PCM-D100 マイクロフォンは OLYMPUS 社行車両がないときに車両をカウントした場合走行車両が製 ME-30W である走行音はサンプリング周波数 48 khz ないときに車両をカウントしなかった場合である TN の 16 bit 長のステレオデータとして録音したまた録音と回数を数えるため車両の通過と車両の通過の間の状態を同時に道路をカメラで撮影しこの映像を真値として利用非検出状態と定めた左右個別の検出結果とそれらを加算 c 016 Information Processing Society of Japan 5

6 1 Table 1 True Positive TP False Negative FN False Positive FP True Negative TN Number of true positives (TP), false negatives (FN), false positives (FP), and true negatives (TN) TP FN FP TN Left to Right Right to Left Total TN TP FN FP TN Accuracy Precision Recall F F-measure Accuracy = TP + TN TP + FP + FN + TN Precision = TP TP + FP Recall = TP TP + FN Precision Recall F measure = Precision + Recall (7) (8) (9) (10) F 4. 1 TP FN FP TN F % % [16] False Negative 1 False Negative DTW 1 5. Dyanamic Time Warping DTW [1] Forren, J. F. and Jaarsma, D.: Traffic Monitoring by Tire Noise, Proc. IEEE Conf. Intelligent Transportation Systems (ITSC), pp (1997). [] Chen, S., Sun, Z. P. and Bridge, B.: Automatic traffic monitoring by intelligent sound detection, Proc. IEEE Conf. Intelligent Transportation Systems (ITSC), pp (1997). [3] Chen, S., Sun, Z. and Bridge, B.: Traffic monitoring using digital sound field mapping, IEEE Trans. Veh. Technol., Vol. 50, No. 6, pp (001). [4] Barbagli, B., Manes, G., Facchini, R. and Manes, A.: Acoustic sensor network for vehicle traffic monitoring, Proc. IEEE Int. Conf. Advances in Vehicular Systems (VEHICULAR), pp. 1 6 (01). [5] ASN pp (016). [6] Vol. 47, No. 8, pp (005). [7] Vol. 15, pp (006). [8] CCTV pp. 1 4 (007). [9] Buch, N., Cracknell, M., Orwell, J. and Velastin, S. A.: Vehicle Localisation and Classification in Urban CCTV Streams, Proc. ITS World Congress, pp. 1 8 (009). [10] Nurhadiyatna, A., Hardjono, B., Wibisono, A., Jatmiko, W. and Mursanto, P.: ITS Information Source: Vehicle Speed Measurement Using Camera as Sensor, Proc. Int. Conf. Advanced Computer Science and Information Systems (ICACSIS), pp (01). [11] Vol. 40, No. 6, pp (005). [1] pp (014). [13] Wu, H., Siegel, M. and Khosla, P.: Vehicle Sound Signature Recognition by Frequency Vector Principal Component Analysis, Proc. IEEE Instrumentation and Measurement Technology Conf. (IMTC), Vol. 1, pp (1998). [14] (01). [15] Knapp, C. H. and Carter, G. C.: The Generalized Correlation Method for Estimation of Time Delay, IEEE Trans. Acoust., Speech, Signal Process., Vol. 4, No. 4, pp (1976). [16] IT (009). c 016 Information Processing Society of Japan 6

IPSJ SIG Technical Report Vol.2018-ITS-72 No /3/ ITS Intelligent Transport System: 2 DTW Dynamic Time Warping RANSAC Random Sampling

IPSJ SIG Technical Report Vol.2018-ITS-72 No /3/ ITS Intelligent Transport System: 2 DTW Dynamic Time Warping RANSAC Random Sampling 1 1 1 2 1 ITS Intelligent Transport System: 2 DTW Dynamic Time Warping RANSAC Random Sampling Consensus 10 F 0.83 RANSAC Random Sampling Consensus Accuracy Improvement of Successive Vehicle Detection in