Research

Georgia Institute of Technology

• Developed model to automated highway maintenance.

• Leverage convoluted neural network using Keras to automatically process drone images to detect areas with mowed grass and areas with unmowed grass, as well damage signs or pavements.

• Obtained average F1 score of 98%, average precision of 99%, and average recall of 96%.

• In collaboration with and funded by the Georgia Department of Transportation.

• Technologies used: UAVs, Python, Keras, OpenCV, C++, CloudCompare, MeshLab.

Georgia Institute of Technology

• Conducted research on accurate and optimum methods to automatically filling occlusions in point cloud data.

• Engineered deep learning model using Tensorflow to detect and predict voxels that appropriately fill in holes.

• Obtained average F1 score of 69%, average voxel precision of 78%, and average recall of 63%.

• Technologies used: Laser scanner, C++, Tensorflow, python, CloudCompare.