Machine Learning Algorithm Development for Footballer’s Performance Sensor Device

Our customer is a Europe-based company developing performance sensors and activity trackers that are used in football training. Their products help in tracking and improving player skills and well-being, as these devices can be worn on and off the field.

The customer was developing a wearable device that tracks football player activity and metrics. They partnered with Embitel as we had prior experience in IoT application development, based on machine learning.
 

• The device is affixed to the leg of the football player while playing the game. Data is collected via the IMU sensor on the leg band.
• IMU sensor data includes 9-axes data of accelerometer, gyroscope and magnetometer.
• We developed the Machine Learning (ML) algorithm to generate minute by minute metrics of the actions of the football player including idle, walk, run, sprint, kick, pass, etc. using IMU sensor data.
• The data is labeled by synchronizing videos and sensor data.​
• Feature Extraction (Average Acceleration, Max & Min Amplitude, Butterworth Filter, etc. apart from common features like magnitude)​ is performed to enhance the quality of the data.
• Time series LSTM model is used to identify various activities like kick, pass, dribble, sprint, walk, run, idle, etc.​
• Metrics (results) include duration of each activity, distance in case of walk, run and sprint, speed in case of kick and sprint, number in case of kick and pass, intensity of each activity, etc.

Challenges faced during the project lifecycle: