該研究的目的是為僅使用地板壓力傳感器數據進行個人識別的腳步識別系統構建深度學習架構。本文首先探討了以時空形式表示獲得的壓力數據的方法。該方法已在最大足跡數據庫（SFootBD）中進行了驗證，以構建強大的特徵學習神經網絡體系結構。根據可用於模型訓練的足跡數據的數量（從最大的訓練數據集到最小的訓練數據集），數據庫將數據庫分為三個關鍵的數據驅動的安全方案。在第二階段中，使用端到端可訓練的3D卷積神經網絡構建一種用於自動特徵提取的新穎體系結構，以學習隱藏的空間和時間特徵。在構建用於特徵提取的神經網絡時，要考慮三個主要限制條件：1）淺層神經網絡以避免過度擬合； 2）從僅地面壓力傳感器提取的低分辨率圖像中進行特徵提取； 3）用於模型訓練的數據集的可用性較低。實驗表明，基於高級語義特徵的時態表示學習更加有用。因此，構建了淺層神經網絡體系結構，以學習神經網絡開始時的空間特徵和神經網絡後期的時間特徵。當與其他具有成本效益的設計結合使用時，與最新方法相比，該系統可有效識別足跡。當訓練數據集很大時，系統可達到98％的分類率，而當訓練數據集最低時則可達到90％的分類率。

The purpose of the study is to build a deep learning architecture for a footstep identification system using only the floor-only pressure sensor data for personal identification. This thesis first examines the way to represent the obtained pressure data in the Spatio-temporal form. The methodology is validated in the largest footstep database (SFootBD) to build a robust feature learning neural network architecture. The database is organized into three critical data-driven security scenarios, according to the quantity of footstep data made available for model training (largest training dataset to smallest training dataset). In the second stage, a novel architecture is built for automatic feature extraction using end-to-end trainable 3D Convolutional Neural Network to learn both the hidden spatial and temporal features. Three main constraints are considered while building the neural network for feature extraction 1) shallow neural network to avoid overfitting, 2) feature extraction from low-resolution images extracted from floor-only pressure sensors, 3) less availability of dataset for model training. The experiments suggested that the temporal representation learning on high-level semantic features are more useful. Thus, the shallow neural network architecture is built to learn spatial features at the beginning and temporal features at the later stages of the neural network. When combined with other cost-effective designs, the system results in effective footstep identification compared to the state-of-the-art approaches. The system achieves a 98% classification rate when the training dataset is large and achieves a 90% classification rate when the training dataset is lowest.

[1] A. Itai and H. Yasukawa, “Footstep Classification Using Simple Speech Recognition Technique,” Proc. IEEE Int’l Symp. Circuits and Systems, pp. 3234-3237, 2008.
[2] L. Middleton, A.A. Buss, A.I. Bazin, and M.S. Nixon, “A Floor Sensor System for Gait Recognition,” Proc. IEEE Workshop
Automatic Identification Advanced Technologies, pp. 171-176, 2005.
[3] G. Qian, J. Zhang, and A. Kidane, “People Identification Using Floor Pressure Sensing and Analysis,” IEEE Sensors J., vol. 10,
no. 9, pp. 1447-1460, Sept. 2010.
[4] J. Yun, “User Identification Using Gait Patterns on UbiFloorii,”Sensors, vol. 11, no. 3, pp. 2611-2639, 2011.
[5] M.S. Nixon, T.N. Tan, and R. Chellappa, Human Identification Based on Gait. Springer, 2005.
[6] R. Vera-Rodriguez, P. Tome, J. Fierrez, and J. Ortega-Garcia,“Fusion of Footsteps and Face Biometrics on an Unsupervised and
Uncontrolled Environment,” Proc. SPIE Biometric Technology forHuman Identification IX, 2012.
[7] R. Vera-Rodriguez, N.W.D. Evans, and J.S.D. Mason, “Footstep Recognition,” Encyclopedia of Biometrics, pp. 550-557, Springer,
2009.
[8] A. Pedotti, “Simple Equipment Used in Clinical Practice for Evaluation of Locomotion,” IEEE Trans. Biomedical Eng., vol. 24,
no. 5, pp. 456-461, Sept. 1977.
[9] M.D. Addlesee, A. Jones, F. Livesey, and F. Samaria, “The ORL Active Floor,” IEEE Personal Comm., vol. 4, pp. 35-41, Oct. 1997.
[10] J. Suutala and J. Roning, “Methods for Person Identification on a Pressure-Sensitive Floor: Experiments with Multiple Classifiers and Reject Option,” Information Fusion, vol. 9, pp. 21-40, 2008.
[11] James Eric Mason • Issa Traoré Isaac Woungang, “Machine Learning Techniques for Gait Biometric Recognition”
[12] Omar Costilla-Reyes, Ruben Vera-Rodriguez, Patricia Scully, “Analysis of Spatio-temporal Representations for Robust Footstep Recognition with Deep Residual Neural Networks”, IEEE transactions on pattern analysis and machine intelligence 2018
[13] Patrick Connor, Arun Rossb, “Biometric recognition by gait: A survey of modalities and features”
[14] Murray, M.P., 1967. Gait as a total pattern of movement: Including a bibliography on gait.
Am. J. Phys. Med. Rehab. 46 (1), 290–333.
[15] Orr, R.J., Abowd, G.D., 2000. The smart floor: a mechanism for natural user identification
and tracking. Extended Abstracts on Human Factors in Computing Systems (CHI’00).
pp. 275–276.
[16] Little, J., Boyd, J., 1995. Describing motion for recognition. Proceedings of the 1995
International Symposium on Computer Vision. pp. 235–240.
[17] Huang, P.S., Harris, C.J., Nixon, M.S., 1999. Recognising humans by gait via parametric
canonical space. Artif. Intell. Eng. 13 (4), 359–366.
[18] Orr, R.J., Abowd, G.D., 2000. The smart floor: a mechanism for natural user identification
and tracking. Extended Abstracts on Human Factors in Computing Systems (CHI’00).
pp. 275–276.
[19] Kumar, V., Ramakrishnan, M., 2011. Legacy of footprints recognition–a review. Int. J.
Comput. Appl. 35 (11), 9–16.
[20] Kuragano, T., Yamaguchi, A., Furukawa, S., 2005. A method to measure foot print similarity for gait analysis. Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation and the International Conference on Intelligent Agents, Web Technologies and Internet Commerce. vol. 2. pp. 816–822.
[21] Orr, R.J., Abowd, G.D., 2000. The smart floor: a mechanism for natural user identification
and tracking. Extended Abstracts on Human Factors in Computing Systems (CHI’00).
pp. 275–276.
[22] Suutala, J., Röning, J., 2004. Towards the adaptive identification of walkers: automated feature selection of footsteps using distinction sensitive LVQ. International Workshop on Processing Sensory Information for Proactive Systems (PSIPS 2004). pp. 14–15.
[23] Rodriguez, R.V., Evans, N.W.D., Mason, J.S.D., 2009. Footstep recognition. Encyclopedia
of Biometrics. Springer, pp. 550–557.
[24] Addlesee, M.D., Jones, A., Livesey, F., Samaria, F., 1997. The ORL active floor. IEEE Pers. Commun. 4, 35–41.
[25] Derlatka, M., 2013. Modified kNN algorithm for improved recognition accuracy of biometrics system based on gait. Computer Information Systems and Industrial Management. Springer, pp. 59–66.
[26] Moustakidis, S.P., Theocharis, J.B., Giakas, G., 2009. Feature extraction based on a fuzzy
complementary criterion for gait recognition using GRF signals.
[27] Nakajima, K., Mizukami, Y., Tanaka, K., Tamura, T., 2000. Footprint-based personal recognition. IEEE Trans. Biomed. Eng. 47 (11), 1534–1537.
[28] Morishita, H., Fukui, R., Sato, T., 2002. High resolution pressure sensor distributed floor
for future human-robot symbiosis environments. Proceedings of the 2002 IEEE/RSJ International Conference on Intelligent Robots and Systems. vol. 2. pp. 1246–1251.
[29] Middleton, L., Buss, A.A., Bazin, A., Nixon, M.S., 2005. A floor sensor system for gait
recognition. Proceedings of the Fourth IEEE Workshop on Automatic Identification Advanced Technologies. pp. 171–176.
[30] Yun, J., Lee, S., Woo, W., Ryu, J., 2003. The user identification system using walking
pattern over the ubiFloor. Proceedings of International Conference on Control,
Automation, and Systems. vol. 1046. pp. 1050.
[31] Yun, J., Woo, W., Ryu, J., 2005. User identification using users walking pattern over the
ubiFloorII. Computational Intelligence and Security. Springer, pp. 949–956.
[32] Qian, G., Zhang, J., Kidané, A., 2010. People identification using floor pressure sensing and analysis. IEEE Sens. J. 10 (9), 1447–1460.
[33] Pataky, T.C., Mu, T., Bosch, K., Rosenbaum, D., Goulermas, J.Y., 2011. Gait recognition:
highly unique dynamic plantar pressure patterns among 104 individuals. J. R. Soc. Interface.
[34] Connor, P.C., 2015. Comparing and combining underfoot pressure features for shod and unshod gait biometrics. Proceedings of the 2015 IEEE International Conference on Technologies for Homeland Security (HST).