Ruchita Vehale

Designed and deployed image segmentation pipelines to precisely isolate defect regions across complex vehicle surfaces, improving detection accuracy and repeatability in production.

Built a Blender plugin automating 3D model processing workflows. A 2-day manual task compressed to under 1 hour (16× improvement), deployed in live inspection pipelines.

Implemented 3D model + 2D image hybrid fusion for pose estimation, significantly reducing reliance on expensive LiDAR hardware while maintaining accuracy across inspection stations.

Developed calibration and localisation algorithms across multiple inspection pipelines, reducing manual intervention and enabling scalable deployment.

Improved codebase architecture and version control workflows, enabling faster, more reliable production deployments across the R&D team.

Engineered a ground-based autonomous precision landing system using an upward-facing Raspberry Pi Camera on the landing pad; ran YOLOv8 inference to detect incoming drones and streamed corrective MAVLink LANDING_TARGET messages at 10 Hz to a Cube Orange flight controller over Wi-Fi UDP, enabling closed-loop autonomous landing without GPS.

Implemented a custom DeepSORT multi-object tracker from scratch — Kalman filter state prediction, Hungarian algorithm for detection-to-track assignment, and CNN appearance embeddings for re-identification — integrated with a stereo vision depth pipeline (stereo rectification + block matching) to compute metric 3D drone position from disparity maps.

Built a pluggable detection architecture (motion, brightness, YOLOv8, YOLOv8+DeepSORT) with an ESP32 Bluetooth communication module for real-time telemetry transfer, Feetech STS3032 smart servo integration via 10 ArduPilot Lua scripts for servo control and health monitoring, and a Flask web interface with live MJPEG stream and browser-based camera calibration.

Contributed to R&D and assembly of a coaxial drone platform for synchronised lighting displays, covering mechanical integration, ESC calibration, flight controller configuration, and structured performance testing to validate flight stability and payload capacity.

Prototyped a precision autonomous landing system fusing GPS position data with camera-based visual detection to improve landing accuracy and repeatability in dynamic outdoor environments.

Optimised antenna designs using CST Microwave Studio, achieving 20% size reduction and 70% accuracy improvement for GPS, Wi-Fi, Bluetooth and 5G frequencies.

Operated 3D printers (FDM, Inkjet) and characterisation tools (XRD, VNA) for antenna prototyping and dielectric material testing.

Fabricated a biodegradable-ink strain sensor via 3D inkjet printing, contributing to sustainable flexible electronics with defence and wearable applications.

Research led to 3 journal publications in Wiley and RSC journals, including novel dielectric nanocomposite materials for microwave applications.

Designed and deployed a MATLAB-based video tracking system using Gaussian Mixture Model and point tracking for moving object recognition in defence research contexts.

Achieved consistent accuracy under variable lighting conditions through adaptive background subtraction and GMM model tuning.

Executed feature extraction and depth estimation using projective geometry across multiple camera planes for accurate 3D reconstruction.

Validated precise velocity measurement across 20 independent video datasets, with validated field performance.