Sports Activity Recognition with Minimal Sensor Data: A New Approach

Imagine having the ability to identify sports activities just by using basic motion data. This is exactly what a new method called Human Activity Recognition (HAR) does. By using sensors and special learning algorithms, HAR can tell the difference between various sports activities with impressive accuracy. This helps everyone, from kids to adults. The future looks bright where AI can tackle tough problems in science. But how does this work? It’s all about using motion sensor data and some clever models called Convolutional Neural Networks (CNN). The process involves four main steps. First, the data from the sensors is cleaned up and made uniform. Then, special tools like Discrete Wavelet Transform (DWT) and Short-Time Fourier Transform (STFT) are used to figure out the key features of the signals.

Next, two CNN models get to work. Each model builds up a picture of the motion features based on the sensor data. When these pictures are combined, a Random Forest classification model steps in. This model has a good look at the features and can tell what kind of sport activity is happening. To check how well this method works, researchers used a dataset called DSADS. The results were amazing – the method could identify different sports activities with a mean precision of 99. 61%.

Actions