- SONY: SSE-TN1W
- Zepp Tennis 2
- Accurate Recognition of Player Identity and Stroke Performance in Table Tennis Using a Smart Wristband
- Action Recognition and Application of Table Tennis Training Based on IOT Perception
- An Approach to 3D Gyro Sensor Based Motion Analysis in Tennis Forehand Stroke
- An_embedded_6-axis_sensor_based_recognition_for_tennis_stroke
- Tennis Stroke Classification: ComparingWrist and Racket as IMU Sensor Position
7 Sensores de Gryo establecidos en diferentes puntos do corpo. A velocidade angular obtida polos sensores xiróscopos debe ser transformada en ángulos de roll-pitch-yaw.
Microcontrolador de 6 eixes JY-61. Modulo de control STM32F405 por BLuettoth. Instalado en raqueta, realtime a mobil. (98% golpes, 96% efectos)
3 etapas:
- Reconocemento de golpe (fluctuación de aceleración en ventanas).
- Clasificación de golpe (forehand, backhand, saque) por aceleración e velocidad angular.
- TopSpin o BackSpin rotación de pelota por velocidad angular.
Comparativa de aceleración maxima, desviación estandar e diferencia entre maximo e minimo. O mellor foi desviación estandar.
Smash e saque valores importantes en Y. Para diferenciar saque de smash, para o saque vai existir unha gran difrencia temporal entre o saque e o ultimo golpe ao contrario que no smash.
MIRAR REFERENCIAS INTERESANTES
Compara muñeca con raqueta. Para clasificar os tipos de golpes usa 4 modelos de machine learning.
Métodos para detectar golpes:
- Acelaración pico con threshold: 9g en x axis
- Calculated accelaration de 3 o 8 gs
- Falsos negativos fora con gaps de 1.25 segundos ou or alternatively gyroscope resultant values within 0.06 s on both sides of the acceleration maxima exceeding ±400 deg.
Machine Learning usados Naive Bayes , support vector machine (SVM), decision and classification tree(CT) random forest (RF) k-nearest neighbor (KNN), neural network (NN) and logistic regression(LG) Further models include AdaBoost and discriminant analysis.
Usan acceleration and angular velocity
Mellores resultados SVM linear con datos brutos de acelaración e giroscopio pasada por un PCA.
- L. Chen, J. Hoey, C. D. Nugent, D. J. Cook and Z. Yu, "Sensor-Based Activity Recognition," in IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), vol. 42, no. 6, pp. 790- 808, Nov. 2012.
- K. F. Li, A. M. Sevcenco and K. Takano, "Real-Time Classification of Sports Movement Using Adaptive Clustering," Complex, Intelligent and Software Intensive Systems (CISIS), 2012 Sixth International Conference on, Palermo, 2012, pp. 68-75.
- D. Conaghan, P. Kelly, and N. E. O’Connor. Game, shot and match: Event-based indexing of tennis. In 9th International Workshop on Content-Based Multimedia Indexing, pages 96–102, 2011.
- D. Connaghan, S. Hughes, G. May, K. O’Brien, P. Kelly, C. ´O Conaire, N. E. O’Connor, D. O’Gorman, G. Warrington, A. F. Smeaton, and N. Moyna. A sensing platform for physiological and contextual feedback to tennis athletes. In Body Sensor Networks (BSN) Workshop, pages 224– 229, 2009.
- D. Connaghan, P. Kelly, et al., “Multi-sensor classification of tennis strokes,” IEEE Sensors, pp. 1437–1440, 2011.
- R. Srivastava, A. Patwari, et al., “Efficient Characterization of Tennis Shots and Game Analysis using Wearable Sensors Data”, IEEE Sensors, pp. 1–4, 2015.
- http://www.st.com/content/ccc/resource/technical/document/datasheet/ef/92/76/6d/bb/c2/4f/f7/DM00037051.pdf/files/DM00037051.pdf/jcr:content/translations/en.DM00037051.pdf
- IMU Data Collection (Python on Android) → Using BLE (Bluetooth Low Energy), you collect IMU data on your mobile device.
- Send Data to Cloud → Use Firebase, AWS, or a REST API to send data to a cloud database.
- Process Data in the Cloud → Apply ML models or signal processing (on Google Cloud Functions, AWS Lambda, or a Python server).
- Store Processed Data → Save results in a database (Firebase, PostgreSQL, etc.).
- Retrieve Data in an Android App → The app fetches processed data from the cloud.