Breaking Down Motion Processing

Ever wonder what makes a motion product "work"? In this blog post, we've distilled motion processing down into digestible bits for anyone to understand. This is truly where the magic happens for Kiwi.


The three types of motion processing that occur are:


1. Rep & Peak Counting

For every type of movement there is a simple threshold and interval between peaks that occurs, this is how we can count how many times a specific movement occurs. For example, in step counting, it's simple peak detection.


Running Peak DetectionFrom this running data, there are very clear peaks to classify a motions repetitions



2. Recognizing New Motions

This occurs by labelling data as it comes in to match a specific movement. From here two data forms are created - both have been used heavily in speech recognition for the past 30 years:





GMM in action, part of Kiwi's internal tools to help with our processes



3. Sensor Fusion

Sensor fusion occurs when you apply math that has existed for decades: quaternion and displacement filters. These allow you to convert Inertial Measurement Unit (IMU) data into 3D directions that can be used in a visual sense, drones, video games, feedback in a mobile app - you name it. Although extremely useful,  there exists a challenge with sensor fusion when adjusting for each IMU nuance.


If you're interested in trying a real-time implementation of motion processing, feel free to give us a shout via our contact form!

John David Chibuk

Toronto, Canada |

Entrepreneur in machine learning and sensor based software development (>10 years). Background in engineering, tech with professional experience in North America and Europe.