Case Study
ML-based motion detection and classification system for a boxing workout product
About the client
Boxx+ is a startup offering a boxing-inspired fitness app connected to wearable Punch Pods that helps users improve their boxing technique. Bluetooth trackers detect a punch, classify its type, and provide exercise tracking to enable users to master their technique.
The challenge
The client was looking to build a boxing solution with Bluetooth trackers (punch pods) connected to a mobile app for end users. The system needed to detect a type of punch (jab, cross, hook, uppercut) in two different stances (Southpaw and Orthodox) to help non-professional and professional boxers track and improve their technique. The client expected the final delivery to be in the form of SDKs that can be easily integrated into native iOS and Android platforms.
Delivered value
We built a high-precision model that can detect and classify mid-air punches and strikes into the punching bag. It was adapted for boxers with any training level, which means they don’t have to provide any additional parameters to track their boxing activity. Since the model we built was universal, there was no need to create separate models for several punch types.
About the project
We started the project by asking the client to structure their requirements and list specific boxing punches they want the system to detect. As a result, we received the requirement to build a model that could determine several punches like jab, cross, hook, and uppercut in two stances: Southpaw and Orthodox.
Based on the client's initial requirements, we conducted an estimation and started working. Our main task was to create SDKs and libraries that communicate with Bluetooth trackers to detect and classify the type of punch with the help of an algorithm.
Initially, we had to collect a dataset to build a model that would detect each type of punch.
We analyzed a selection of different participants performing all types of punches. Then, we moved on to the next stage — training the model based on the collected data and implementing the model into SDKs.
As a result of the collaboration, we provided the client with a model, enabling users to track their boxing activity. It also detects versatile punch types performed by boxers with any training level.
Technologies
C++
Python
SGTM neural networks library
Android SDK
iOS SDK
Bluetooth Low Energy (BLE)
Save case study in PDF
How it works
AS
AS
We have thoroughly enjoyed working with Lemberg Solutions. They are very efficient and provide accurate timelines for their work. We are really pleased with the project outcome and look forward to working with them again soon.
Download the case study to read it later