Flooding Level Classification by Gait Analysis of Smartphone Sensor Data
Published in IEEE Access, Volume 7 (December 2019)
DOI: 10.1109/ACCESS.2019.2959557
Citation: Ujjawal K. Panchal, Hardik Ajmani, and Saad Y. Sait
Co-authored research paper published under the guidance of Prof. Saad Yunus Sait. This work presents a novel, low-cost approach to urban flood detection—a critical infrastructure challenge affecting Chennai and cities across India.
Problem Statement
Urban flooding in India causes:
- Significant casualties annually
- Financial losses exceeding tens of billions of rupees
- Inefficient emergency response due to lack of real-time flood level data
- Poor urban planning for storm water management
Traditional satellite-based flood detection (Synthetic Aperture Radar) fails in densely populated urban areas where buildings and infrastructure obstruct sensor readings. We needed a scalable, cost-effective solution.
Solution: Gait-Based Flood Detection
Core Innovation
We discovered that human gait characteristics change distinctly at different water depths. By leveraging ubiquitous smartphone sensors (accelerometer, gyroscope, magnetometer), we could classify flooding levels without any additional infrastructure.
Methodology
Data Collection:
- 12 volunteers walked in pools at varying depths (simulating different flood levels)
- Smartphone sensor data captured using AndroSensor app
- Time-domain features (mean, variance, etc.) and frequency-domain features (FFT coefficients) extracted
Machine Learning Models Tested:
- Support Vector Machines (SVM)
- Random Forests
- Naïve Bayes
Results
Support Vector Machines achieved 99.45% classification accuracy, significantly outperforming competing models.
Feature analysis confirmed our intuition—gait characteristics (stride length, movement patterns, acceleration) vary predictably with water depth, enabling reliable flood level detection.
Real-World Impact
This research enables:
- Zero-cost deployment: Leverages existing smartphones already in use
- Real-time flood mapping: Users open an app while walking through flooded areas, which records sensor data and classifies flooding level
- Scalable emergency response: Central server aggregates location and flood level data for rapid dispatch
- Infrastructure planning: Data informs construction of appropriate storm water drains to mitigate future flood damage
Publication Details
- Journal: IEEE Access
- Volume: 7
- Pages: 181678 - 181687
- Publication Date: December 13, 2019
- License: Creative Commons Attribution 4.0 (Open Access)
- Impact: 5+ citations in peer-reviewed papers
Key Contributions
- Novel technique for flood detection requiring no specialized equipment
- Practical solution addressing a real-world problem affecting millions in India
- Validated machine learning approach with 99.45% accuracy
- Scalable framework leveraging ubiquitous smartphone technology
Technologies & Skills
Machine Learning · Gait Analysis · Smartphone Sensors · Accelerometer Data · Feature Engineering · Support Vector Machines · Python · Time-Series Analysis · Signal Processing · Data Classification
Read the full paper: IEEE Access - Flooding Level Classification
Mentor: Prof. Saad Yunus Sait
Research Focus: Urban Computing & Flood Disaster Mitigation