A Computer Vision-Based System for Stride Length Estimation using a Mobile Phone Camera
Wei Zhu, Boyd Anderson, Shenggao Zhu, Ye Wang · 2016 · ASSETS '16: Proceedings of the 18th International ACM SIGACCESS Conference on Computers and Accessibility · doi:10.1145/2982142.2982156
Summary
This paper presents an affordable computer vision system for measuring stride length in Parkinson's disease (PD) patients using only a smartphone camera and a printed PVC mat. Gait analysis is a critical non-invasive method for PD diagnosis and assessment, but clinical systems like GAITRite (pressure-sensitive mats) or Vicon (motion capture cameras) cost $10,000-$45,000 and require specialized lab settings, making them inaccessible for rural hospitals, developing countries, and home monitoring. The researchers designed a simple system with three components: a 0.9m x 5.0m PVC mat with alternating black and white markers printed along its edges, a smartphone mounted on a tripod to record walking videos, and computer vision algorithms to process the footage. The mat's markers provide reference points for calculating scale and perspective. The system pipeline includes mat extraction (detecting markers and determining camera orientation), shoe detection (identifying foot contours using color segmentation and edge detection), and stride length estimation (mapping pixel positions to real-world measurements using perspective-aware algorithms). A key innovation is handling perspective distortion from the camera angle—the researchers developed multiple mapping functions that account for the non-parallel marker lines caused by camera perspective, achieving more accurate distance calculations than simple linear approaches.
Key findings
The system achieved a mean absolute error of 0.62 cm for stride length measurement, comparable to the "gold standard" GAITRite system. In lab testing with 128 video sessions containing 382 strides, participants simulated PD gait patterns including start hesitation, asymmetric walking, and freezing of gait. Clinical validation at Huashan Hospital in Shanghai involved 55 elderly subjects (44 PD patients across four severity stages and 11 healthy controls) over two weeks, producing 98 videos with 1,947 walking sessions. The system correctly identified every shoe contour, counted every stride accurately even for patients with abnormal gait patterns, and successfully extracted the walking mat in all videos across varying lighting conditions, clothing types, and gait abnormalities. The total system cost is under $800 USD: $200-$700 for a smartphone, $50-$80 for the printed mat, and $20 for a tripod—orders of magnitude cheaper than commercial alternatives. The system generates statistical reports including mean stride length, stride time, coefficient of variation, and standard deviation for each foot.
Relevance
This research addresses a critical healthcare accessibility gap: neurological assessment tools that are prohibitively expensive for most clinical settings worldwide. For people with Parkinson's disease, regular gait monitoring is essential for tracking disease progression and adjusting treatment, but access to clinical gait analysis is limited outside major medical centers. The smartphone-based approach enables gait assessment in rural hospitals, community clinics, and potentially patients' homes—democratizing access to objective mobility measurement. The system's portability and low cost make it particularly valuable for developing countries where specialized medical equipment is scarce. For accessibility practitioners, this demonstrates how consumer technology and clever algorithmic design can replicate expensive clinical capabilities at a fraction of the cost. The principles could extend to other movement disorder assessments or physical rehabilitation monitoring, expanding accessible healthcare beyond traditional clinical settings.
Tags: Parkinson's disease · gait analysis · computer vision · mobile health · movement disorders · neurodegenerative diseases · low-cost technology · healthcare accessibility · motor impairment