Gait analysis is crucial for evaluating neurological disorders, such as Normal Pressure Hydrocephalus (NPH), Parkinson's disease, and Multiple Sclerosis. Current methods often require clinic visits and trained personnel, limiting accessibility. 

This study introduces a mobile phone application designed to assess gait performance using the 30-second sit-to-stand test, 10 Meter Walk test, and Timed Up and Go (TUG) tests. The app employs advanced computer vision techniques, using BlazePose 3D algorithms to extract key points on the human body, which then will be classified into different positions using dense neural network. These classifications then will be used to measure patients' gait performance.

The app processes real-time camera feeds at an average of 18 frames per second, tracking 3D key points from 33 body parts with an 80% accuracy. It achieved 98% accuracy in classifying patients positions into sitting, standing, and falling. The app was tested on 40 sit-to-stand and 40 timed-up-and-go (TUG) tests. In sit-to-stand assessments, the app's measurements were perfectly correlated with human evaluations (r = 1.00, P = 1.00). For TUG tests, the app accurately assessed patients in 38 out of 40 cases (tracking failed in 2 tests), with a high correlation to human measurements (r = 0.99, P < 0.0001).

This mobile application demonstrates high accuracy in classifying positions and performs reliably in gait assessments. It provides a practical, accessible solution for remote gait analysis, potentially reducing the need for hospital visits. Initial testing shows great promise in enhancing patient care through convenient, real-world gait assessments.