Realtime Pothole Detection

Research paper is available here: google scholar, arxiv

Idea

• Mobile phone centric detection
• With accelerometer and gyroscope data
• 50Hz, 20ms sensor refresh rate

Data Acquisition

Format: Time-stamp, Accel X, Accel y, Accel Z, Gyro X, Gyro y, Gyro z

Data modeling

Accelerometer data plot Raw data plot, Mean, Standard Deviation:

Gyroscope data plot Raw data plot, Mean, Standard Deviation:

Confusion matrix:

Export model from MATLAB to C code

Export trained model to C code using MATLAB coder SVM Model and C Code generation repo is available here: https://github.com/aswathselvam/pothole_MATLAB

Import model C code in Android Studio

Use JNI to access the svmPredict() function

TODO

Calibration for running the trained model based on initial orientation of the smartphone.

Calibration shall provide a rotation matrix for coordinate transformation between the orientation used while obtaining the data and the orientaiton of the phone during the detection.

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update tests as appropriate.

Presentation slides: https://docs.google.com/presentation/d/1QHmCzykV8gG-gyEe1Cde11WV_VeJjogQ9z2-AZNWqvg/edit#slide=id.p