Europe Union

At-Home Rehabilitation and Physical Therapy Scaled Up with Computer Vision Engineering

Learn how we used a mobile application equipped with pose detection to help our client’s users perform exercises at home correctly.

Our Partner Is a Health Solutions Startup from Norway

Kinetech is a Norwegian startup focusing on modern well-being and digital innovations that help everyday users with sports activities, rehabilitation, and exercises. Their company builds accessible, high-quality smartphone apps that help users exercise correctly. This time, they needed computer vision and mobile development experts to help them with their next project.

Rehabilitation in Specialised Facilities Isn’t Accessible to Everyone

Our client noticed a problem that plagues society in Norway and other European countries: limited access to professional rehabilitation for those who need it. It can be especially difficult for people with limited mobility. Patients often face long wait times due to the overloaded healthcare system. Thus, our client came up with the idea of creating an advanced, multi-component application that would track the user’s pose, assess the exercise technique and point out the mistakes so they can correct them. 

The application is mainly dedicated to those who need physiotherapy to exercise at home. However, ultimately, our client’s goal is to make it accessible for all smartphone users who want to exercise at home and ensure their technique is correct.

The Key to Build The Application Was Our Expertise in Computer Vision

Kinetech had a clear vision of the product it wanted to create. The missing ingredient was deep technological expertise, which would allow it to develop reliable and accurate pose tracking.

The vision component was the most challenging part. Our client required a strong team of experts in computer vision, signal processing, kinematics, mechanics, and anthropometry to build a mobile application that assists in correctly executing exercises. Since exercise technique is essential, as incorrect execution may lead to injuries or strains, the module had to be accurate and reliable.

The Discovery Process Helped Us Establish the Basics and Outline the Cooperation

When the client reached out to us, they already knew how we worked and what they could expect since we successfully collaborated on a different project, which made building mutual trust easier. We spoke to the company’s three founders and built the team that would take over the technological aspect of the product.

During the discovery workshops, we established the product’s expectations and roadmap. In essence, the application should use the phone’s camera to analyse the person’s pose during the exercise in real-time. Then, by examining the angles and position, guide them to the correct exercise technique, for example, squats.

After establishing the goals, budget, and workflow, our work started, and the project is still continuously developed and perfected.

Our Team of Tech Experts Cooperates with Human Body Experts to Create an Accurate Pose Detection Module

Our team included a Project Manager to coordinate the work and communication, a Senior Data Scientist and two UX and UI experts who took over the app interface to make it clean and user-friendly. However, since the human body is highly complex, we combined forces with company executives and the Norwegian School of Sport Sciences and Muscle Animations.

Their aid is invaluable to us, as it enables us to understand the movements and helps us ensure we can build our models and algorithms as accurately as possible, primarily since we aim to add 3D depth to the detection made by only a smartphone camera. Ultimately, we must ensure the application works as intended and correctly guides the users through the exercises to avoid harm.

Computer Vision Pose Detection for Rehabilitation

Estimates Confronted with Real-Life Examples to Fine-Tune the Pose Detection Process

To find the best angle a phone camera can offer to detect the user’s pose accurately, our team has decided to use recordings of people performing the exercise (in our case, squats) to determine the critical points of correctly performed technique. It allowed us to create estimations, which the application would use to guide the user towards the correct position.

Computer Vision Pose Detection

Testing Different Technologies Allowed Us to Find the Best Options for Accessibility and Accuracy

When choosing the technologies most appropriate for accurate pose detection, we had to consider that we would be working with phone cameras, which have their limitations. After testing different options, we made the best decisions to make it precise and accessible.

We’ve Selected Google MediaPipe for Pose Detection and Perfected It to Work with Only a Phone Camera

Finding the best technology to accommodate a single smartphone camera for detecting and assessing the body pose was challenging. After examining different options, we decided to employ Google’s MediaPipe, which proved helpful in reviewing the pose. 

We noticed that the algorithm struggled to stabilise the critical points on the body while performing squats, resulting in visual stuttering. Our experts modified the lines and points to be more stable and show the correct pose outline and assessment. Our team proposed a camera calibration step before performing the exercise to enhance the pose detection process further.

Python Combined with the OpenCV library Enabled Building a Reliable Backend

OpenCV is a robust Python library for computer vision. Its capabilities in object detection helped us build a solid base for pose-detecting models and mechanisms. The language’s flexibility in working with different data types and structures allowed us to adjust the model to the environment of a mobile phone camera.

Our Expertise in Computer Vision and UX Design Allowed Us to Build a Solid Base for Further Development

 

Our team proposed a solution comprising a complex platform with several components. The task involved designing the app’s core, the patient interface, and a pose detection part. 

The project is still in progress within the scope of research and development. We are building a unique solution that breaks the technological state of the art in multiple ways. Due to the project’s innovative nature, it needs time to progress.

We intend to focus on correct camera calibration in the future and improving the detection accuracy. We have a bigger picture ahead of us and have a consistent vision aligned with our clients.

Computer Vision Powered Mobile App for Exercise

See how we built a ball trajectory-tracking app for football enthusiasts.

Read the Case Study

Computer Vision Research to Develop the Final Version of Accurate Pose Detection

We are looking forward to further cooperation and progression as the project grows. So far, the client is highly satisfied with our work and the improvements we’ve achieved with the existing technologies. We’re looking for further breakthroughs and development on both ends.

We are viewing the product and its potential with optimism, anticipating the benefits it will offer once it is launched and available for use.

 See how our computer vision expertise can transform your business.

Computer Vision Services

Read More of Our Case Studies.

Let’s connect!

Send us an e-mail: [email protected]