Accelerometer and machine learning type of motion detection

Question

I'm going to start developing a mobile application that uses an accelerometer to determine the type of movement, whether it be jumping, running, walking, etc. Now I have done a lot of online searches recently and understand a little that using machine learning can detect samples in samples collected from an accelerometer. I am completely unfamiliar with machine learning, but if I understand correctly (based on hints provided by other specialists on the Internet), I can use either Vector Vector Machines or Neural Networks to recognize patterns in the collected patterns and match them with a specific type of movement . I also know that the OpenCV library provides both methods.

Can someone who has experience in the field tell me which method is better to use, and also help me complete the steps necessary to collect data before presenting the result?

Answer 1

There is no convincing evidence that SVM or NN is better for the general case, and the performance is highly dependent on the application and how you configured both algorithms. Thus, the only way to find out about any new applications is to try both of them using the same data and see which one works best.

In addition, NN is usually calculated faster for classification, but slower for training. SVM is faster for training, but slower for classification.

For your case, what input parameters for your algorithm will be more serious. I would not submit raw accelerometer data to SVM or NN. Instead, I would pre-process and get basic information such as total power, standard deviation, and maybe some factors in the frequency domain to determine how fast it is. Performing this method gives you the best intuition to improve and customize your classifier. If you have the correct input parameters, you may not even need NN or SVM to determine the basic movements, a simple identifier of the nearest distance may work. Good luck.