Developed an image-based road surface inspection product from the ground-up. Achieved surface defect measurement accuracy of < 4mm from a monocular road facing video.• Developed pipeline for 3D reconstruction of road surface.• Developed an algorithm for generating depth map and stitched 2D birds-eye-view (BEV) images of road surface.• Implemented planar surface segmentation algorithm for point-clouds to measure depth-based defects.• Trained CNN for semantic segmentation of road assets. Achieved IOU 91.34%.• Fused IMU and GPS data for robust scale estimation• Developed multiple ML and image processing based algorithms for calculating pavement condition index (PCI) based on client requirements.

• Improved existing algorithm for object tracking and inventory in cluttered background by 11% using classical CV and deep learning approaches.

Worked on classification and localization of abnormalities in EEG signals for assisting neurologists in diagnosis:• Developed CWT (continuous wavelet transform) based pre-processing routine to generate scalograms of EEG windows for training CNNs. Led to 25% increase in performance compared to initial baseline.• Trained and evaluated classical and deep learning based classification architectures (GoogLeNet, VGG16, EffienctNetB0, Random Forest Classifiers, Decision Trees, SVM).• Curated and published the first-of-its-kind event localization EEG dataset with a focus on South Asian demographics.