Abnormal Gait Recognition in Real-Time using Recurrent Neural Networks.pdf
- 1. Abnormal Gait Recognition in Real-Time using Recurrent Neural Networks Thanaporn Jinnovart1 , Xiongcai Cai1 , and Kundjanasith Thonglek2 1 University of New South Wales, Sydney, Australia 2 Nara Institute of Science and Technology, Nara, Japan 1
- 2. Outline ❖ Introduction ❖ Methodology ❖ Evaluation ❖ Conclusion 2
- 4. 4 Gait Abnormality ❖ Gait abnormality is a deviation from normal walking (gait) ❖ Watching patient walk is the most important part of the neurological examination ❖ Normal gait requires many systems ➢ Strength ➢ Sensation ➢ Coordination
- 5. 5 Diversity of Gait Abnormalities
- 6. 6 Existing Techniques Doctor’s Diagnosis Motion Sensor Depth Camera
- 8. 8 Approach ❖ We would like to apply artificial intelligence technology to recognize the gait abnormality for increase the model accuracy and decrease the recognize time ❖ We apply the recurrent neural network which is the time series forecasting technique to recognize the gait from the streaming video by using only one RGB camera
- 10. Evaluation 11
- 12. 13 Confusion Matrices Recurrent Neural Network (RNN) Long Short-Term Memory (LSTM) Gated Recurrent Unit (GRU)
- 13. 14 Model Accuracy Method Normal Gait Abnormal Gait Accuracy (%) True False True False HSTOD 728 272 547 453 63.75 RNN 805 195 889 111 84.70 LSTM 887 113 926 74 90.65 GRU 831 169 903 97 86.70
- 14. 15 Training time & Inference time Training Time Inference Time
- 15. Conclusion 16
- 16. Conclusion ❖ The proposed method requires only single RGB camera ❖ The proposed method is able to recognize 5 gait abnomalities ➢ Spastic gait abnormality ➢ Scissors gait abnormality ➢ Steppage gait abnormality ➢ Waddling gait abnormality ➢ Propulsive gait abnormality ❖ The proposed method has the recognition accuracy around 82% 17
- 17. Future Work ❖ We would like to investigate the time-series forecasting techniques other than recurrent neural networks ❖ The walking postures from other large-scale datasets should be used to evaluate the generality of our proposed predictive model for gait abnormalities recognition ❖ Significant features and postures that impact gait abnormalities will be investigated 18
- 18. Q & A Thank you Email: thonglek.kundjanasith.ti7@is.naist.jp 19