Price: 0
Number of applications: 2
22.01.26 (inclusive)
Discussed individually
MVP
ICT tasks
Media sphere
Technologies in transport and logistics
Mobile app
Users massively perform basic exercises with a violation of technique, which: • reduces the training effect; • increases the risk of overload and injury; • Demotivating (slow progress, pain, lack of growth in results). Current situation in the application: • There is activity tracking (approach/repetition counter, manual input); • there is no qualitative analysis of the technique based on video; • The user does not receive objective feedback about the quality of each repetition. Technical limitations and challenges: 1. Video quality and shooting conditions o Different shooting angles (¾, side, front); o Different lighting conditions; o background and noise (other people, objects). 2. Limited computing resources on the device o not all devices pull heavy models in real-time; o it is necessary to provide the ability to perform both on the client (online analysis) and on the server (batch analysis based on uploaded video). 3. Restriction on using only free models/libraries o you cannot use paid APIs (for example, commercial SDKs for motion capture); o you need to rely on open source/free solutions: MediaPipe, OpenCV, PyTorch/TF, ONNX. 4. Formalization of technique into numerical rules o it is necessary to formalize the "correctness" of pull-ups and push-ups into a set of metrics: angles, amplitude, synchronicity; o determine the deviation thresholds at which the system should consider a repeat to be "low-quality" or "partially counted."
Expected user effects: • Increased engagement by: o gamification (assessment of the quality of each approach, progress in technique); o The experience of a "personal trainer" in the app. • Reducing the number of technical errors and complaints of discomfort/pain. • Improvement of subjective satisfaction with the training process. Expected product and business effects: • Retention growth (D7, D30) due to the useful technique analysis function. • Increase the time in the application and the frequency of user returns. • The possibility of monetization: o Advanced technical reports in a paid subscription; o Personalized training plans based on the quality of performance. Expected technological effects: • Accumulation of dataset of real user movements for the subsequent retrained model. • An internal pose analysis platform that can be scaled to other exercises (squats, plank, bench press, etc.).
Iskenderov Erbol
Purpose and description of task (project)
The goal of the project is to implement in a mobile sports application the function of automatic analysis of exercise technique based on user–uploaded videos based on free computer vision models (pose estimation). At the first stage, two basic exercises are considered: • pull-ups on a horizontal bar; • push-ups from the floor. The system must: • accept video from the user (portrait/landscape orientation, side view); • automatically find a person and his supporting projectile (horizontal bar, floor) on the video; • extract the skeletal model (coordinates of joints by frames); • calculate key biomechanical parameters of the equipment; • identify typical errors and the quality level of execution; • give the user a short rating (score) and 2-4 personalized recommendations. Available free pose assessment solutions are used for implementation.: • MediaPipe Pose / BlazePose (Google, free model for 33 skeleton key points); • OpenCV for working with video and basic preprocessing; • Python/Node.js for backend processing; • If necessary, ONNX Runtime / TensorFlow Lite for optimization.
Whatsapp
