Our goal was to track reciprocation exercise movement, and we used WICED Sense to collect the gyroscope and acceleration data. Our first attempt was to use the sensor data to make a full description of the movement, which meant that we wanted to portrait trajectory of the device and further make counting. However, we found that WICED Sense is a relatively “poor” sensor, with which the data collected was full of noise and the acceleration data would steeply fluctuate if the sensor rotates. In this case, the trajectory method would only work for the first and second movement, and then once the acceleration fluctuated, the movement generated by its integral will grow exponentially and never go back again. So, it was impossible to get a full description of the movement, and finally we chose a more “smart” way to detect the movement in gravity direction and got a good result.
After our data cleaning and smoothing processing, we could clearly recognize the patterns. In this project, we have implemented a DSP algorithm for recognizing, and in the future, we want to collect a large dataset and train some machine learning classification models, like SVM, to design a “smart” machine and increase the accuracy of recognition. Furthermore, if it is possible for us to design our own exercise machine equipped with smart sensor, we can change the exercise movement to gear rotation and use gyroscope sensor to detect the times of rotation and further translate it to the distance of movement.
Finally, in our current version app, we have included nine exercises, and made recommendation based on these exercises. We think that this is not enough. In the future, we want to include more exercises in our system, and also we would recommend user to fill in some personal information, like gender, height, weight, etc.. These features can help us to better understand user’s exercise habits and assist us to calculate calories consumed more accurately and in return make more accurate recommendation.
After our data cleaning and smoothing processing, we could clearly recognize the patterns. In this project, we have implemented a DSP algorithm for recognizing, and in the future, we want to collect a large dataset and train some machine learning classification models, like SVM, to design a “smart” machine and increase the accuracy of recognition. Furthermore, if it is possible for us to design our own exercise machine equipped with smart sensor, we can change the exercise movement to gear rotation and use gyroscope sensor to detect the times of rotation and further translate it to the distance of movement.
Finally, in our current version app, we have included nine exercises, and made recommendation based on these exercises. We think that this is not enough. In the future, we want to include more exercises in our system, and also we would recommend user to fill in some personal information, like gender, height, weight, etc.. These features can help us to better understand user’s exercise habits and assist us to calculate calories consumed more accurately and in return make more accurate recommendation.