movAcam case study, motion detection in sport with a video analysis app
How a mobile camera system for sport became a video analysis app with AI motion detection that shows joint points and angles directly in the image. With hosting in Austria.

Making movement visible
A swimmer's elbow enters the water too deep. A runner's knee buckles slightly inwards on landing. Details like these decide tenths of a second and injuries, and they are barely visible to the naked eye. The Austrian company movAcam developed a mobile motion analysis system for exactly this. Together with Anexia Digital Engineering, it has become a video analysis app with AI-assisted motion detection that makes joint points and angles visible directly in the image. This case study describes the task, the approach and what other companies can take away from it.
The client and a system built by sports scientists for practice
The person behind movAcam is MMag. Dr. David Höden, a sports scientist and business economist who earned his doctorate at the University of Graz. He is a state-certified coach for swimming and triathlon and founded the company fluidlife in 2012. He has been developing movAcam since 2020.
The product is aimed at coaches, athletes and therapists. It consists of two synchronised cameras that record at up to 120 frames per second and in resolutions up to 4K, a preconfigured Android tablet, tripods and a transport case. The entire system weighs around four kilograms and, according to the manufacturer, can be set up in less than three minutes. Alongside the version for use on land, there is a version for the water that records swimmers above and below the surface.
movAcam is used in swimming, running and gait analysis, in fitness and rehabilitation, and in ball sports such as football and tennis. Its users include clubs and federations as well as physiotherapists and practices for sports medicine and orthopaedics.
The starting point
Video analysis in sport is nothing new. What is new is the aim of making it so simple that it actually happens in everyday training. If you first have to copy files from two cameras after a session, synchronise them and analyse them on a laptop, in practice you rarely do it.
movAcam solved this problem on the hardware side. Both cameras are plugged in, recording runs in sync, and the video is immediately available on the tablet. What was missing was the next step. Drawing angles by hand takes time and depends on how precisely someone hits the joint point. A video analysis app that recognises the body itself makes analysis faster and easier to compare.
Then there was the question of data. Footage of athletes, including young athletes and patients, is sensitive. A solution that sends footage to a service outside Europe for analysis was out of the question.
The task for Anexia Digital
The collaboration covered three closely interlinked areas. The first was AI motion detection, meaning the automatic recognition of joint points and the calculation of angles in the video. The second was the video analysis app for the Android tablet, used to record, analyse and share. The third covered cloud and hosting for synchronisation, team management and storage in Austria.
Each area would have been a manageable project on its own. The challenge lay in how they work together. Detection that impresses in the lab but takes ten seconds to compute at the poolside helps no coach.
AI motion detection, turning images into skeletons
The technical term for the method behind it is pose estimation. A neural network searches each individual frame for the position of shoulders, elbows, wrists, hips, knees and ankles. Connecting these points produces a stick figure of the body that follows the movement. Joint angles can be calculated from the points and tracked over time.
In the movAcam video analysis app, this skeleton appears as an overlay on the video. Coaches see joint points and angles without having to draw anything themselves, and can concentrate on interpreting what they see.
Why sport is difficult for these methods
Pose estimation models are mostly trained on everyday images. People stand, walk or sit. In sport the world looks different, and four points decide the quality there.
Movements are fast. In a sprint or a serve, arms and legs blur if the camera records too slowly. The high frame rate of the cameras is therefore a prerequisite rather than a luxury, but it also multiplies the number of frames that have to be analysed.
Body parts hide each other. In side view, the rear arm disappears behind the torso. A good system has to handle this without placing points in nonsensical positions.
Postures are unusual. A tumble turn, a bicycle kick or a deep squat rarely appear in ordinary training data.
The environment changes. Indoor lighting, backlight outdoors, reflections and water make very different demands.
Detection for sport must therefore be tested on real footage from the sports hall, the pitch and the pool. The benchmark is whether an experienced coach trusts the result.
Balancing accuracy and speed
The larger a model, the more accurate it usually is and the longer it takes to compute. For a video analysis app used in training, both count. An analysis that is only ready in the evening misses the point, because feedback should come while the athlete can still feel the movement in her body.
The art lies in choosing a model that is accurate enough and whose result is still available on the tablet on site.
The video analysis app on the tablet
A tablet at the poolside or on the sidelines is used differently from a computer in the office. Hands are wet or in gloves, the sun glares, and there are thirty seconds between two runs. The app is designed for these conditions.
After recording, the coach has the tools needed for well-founded feedback. They can use slow motion and frame-by-frame playback, measure angles and work with drawing tools, or overlay two recordings for direct comparison. Still images and markers capture important moments, and a voice commentary is saved with the video. Videos can be trimmed and given an intro and outro, then sent to athletes, fellow coaches or therapists.
Motion detection fits into these tools. It is not a separate mode you have to look for first, it adds another layer to the view.
The app runs on Android. Driving two high-frame-rate cameras at once, showing video smoothly frame by frame and running a neural network alongside requires direct access to the hardware. Our page on Android app development explains why we usually recommend native development for requirements like these.
Cloud and hosting in Austria
An analysis is only worth something once it reaches the athlete. The video analysis app therefore synchronises recordings with a cloud through which coaches manage their teams and members and share videos. Athlete lists can be imported instead of creating each name individually.
The cloud behind the app is operated in Austria. For movAcam this is a selling point, for its users a prerequisite. Clubs work with minors, therapists with patients, and federations do not want the technique videos of their top athletes stored on servers they have no control over.
Here, Anexia contributes its own infrastructure alongside software development. Application, storage and operations come from a single source, and the question of where the data is can be answered in one sentence.
What makes a project like this succeed
Looking back, three things made the difference. They apply far beyond sport.
The client's expertise beats any technology. No developer can judge whether detection is good enough. Only someone who knows what matters in a front crawl stroke or a running stride can do that. Close coordination with movAcam's sports scientists was therefore the most important building block.
Real footage comes before the first model. Anyone who develops with sample videos from the internet will be surprised on site. Material from the sports hall, the pitch and the pool belongs on the table from the start.
The whole counts more than the model. Camera, app, detection and cloud have to work as one product. Users do not distinguish which part happens to be slow.
The result
Today movAcam offers a system with its own video analysis app that combines recording, analysis and sharing in a single workflow. After set-up, recording runs in sync, the analysis happens on site, and videos are edited in the app and sent. Motion detection shows joint points and angles as an overlay on the video, so nobody has to draw them by hand.
What other industries can take away
The technology behind this video analysis app is not limited to sport. The same approach can be applied wherever human movement is assessed.
In physiotherapy and rehabilitation, an app like this makes progress visible, such as the range of motion of a joint over several weeks. In occupational medicine, lifting movements and postures at workplaces can be assessed without attaching sensors to the body. In vocational training, overlaying a reference movement helps people learn manual skills.
The step between people and machines is also smaller than you might think. The same computer vision methods that find a knee joint in a video let robots recognise their surroundings. Our article on robotics development describes how we are building this field right now.
What these applications have in common is the combination of camera, app, AI analysis and operations in which the data stays in Europe. Anexia Digital Engineering delivers software, AI analysis and operations together. Our page on app development gives an overview of our services for mobile applications.
Technology at a glance
Two synchronised cameras serve as the image source, with up to 120 frames per second and resolutions up to 4K. The Android application runs on a tablet with a large screen. Pose estimation detects joint points and calculates angles. A cloud handles synchronisation and the management of teams and members, and hosting takes place in Austria.
Frequently asked questions about the video analysis app with AI
A few questions come up again and again. Here are our answers.
Does a solution like this need special cameras?
For fast movements, yes. Only a high frame rate delivers sharp individual frames on which joint points can be determined reliably. For slower movements such as gait analyses or rehabilitation exercises, the camera of a current tablet or phone is often enough.
How accurate is the automatic angle measurement?
In any video analysis app, that depends on camera position, frame rate, clothing and movement. It is well suited to feedback in training and to comparisons over time. Whether and how measurements may be used in medical diagnosis is a separate question to be clarified with the manufacturer and, if in doubt, legally.
Can we have a video analysis app developed for our own use case?
Yes. Camera, detection and analysis can be transferred to other movements and industries. It starts with the question of what exactly should be measured and how precise it needs to be.
Where are the videos stored?
In the project described here, in Austria. For your own projects, you decide whether the videos are stored in an Anexia Group data centre or in your own environment.
How long does a project like this take?
For a video analysis app, that depends on whether hardware and footage already exist. A first proof of feasibility with real videos is usually possible within a few weeks and shows how much of the way remains.
Your project with moving images and AI
Do you want to make movement visible and measurable, in sport, in therapy or in your operations? In a no-obligation first conversation, we look at your idea and, where available, your footage, and tell you what is possible today with a video analysis app and AI. Use our contact page to arrange a date.


