DRDO — Research Centre Imarat (RCI)Jul 2025 – Sep 2025Hyderabad, India
Responsibilities
Real-time Sign Board Classification and Translation (Python, OpenCV, TensorFlow, YOLO Nano): Built a real-time roadside sign board classification and translation system in Python, using lightweight quantized vision models on edge hardware to cut inference latency to under 15 ms per frame.
OCR & Multi-Class Script Recognition: Engineered OCR and script-recognition pipelines that detect multi-class local scripts with over 98% precision and recall under motion blur, low light, and physical distortion.
Sensor Processing & Trajectory Prediction Simulation (MATLAB, Simulink): Built a multi-sensor signal processing simulation in MATLAB and Simulink that ingests telemetry, radar tracking, and altimeter data across aerospace protocols, reaching sub-meter spatial RMSE in real-time 3D tracking.
Dynamic Target Tracking & Latency Reduction: Implemented machine-learning trajectory prediction for dynamic target tracking, cutting state-estimation latency by 45% while keeping detection probability above 99% under heavy noise.
Hardware Protocol Software (C, RS-422, MIL-STD-1553): Wrote C software that talks to hardware over RS-422 and MIL-STD-1553. Systematic testing and tight hardware-software integration cut communication errors by 35% and improved protocol compliance.
WinAPI Real-Time Telemetry Dashboard: Built a WinAPI dashboard that graphs incoming data in real time, eliminating manual log inspection, and overhauled data handling, synchronization, and error handling in the existing codebase.
Key Achievements
Cut edge vision inference latency to under 15 ms per frame using lightweight quantized YOLO Nano models on constrained hardware
Attained >98% precision and recall across multi-class local script recognition under motion blur, low light, and physical distortion
Reached sub-meter spatial RMSE in real-time 3D tracking by fusing telemetry, radar tracking, and altimeter data in MATLAB & Simulink
Reduced state-estimation latency by 45% with ML trajectory prediction while sustaining >99% detection probability in high-noise environments
Decreased communication errors by 35% and enhanced protocol compliance across RS-422 and MIL-STD-1553 buses
Eliminated manual log inspection for engineering teams by building a responsive WinAPI real-time data graphing interface
Technologies Used
PythonOpenCVTensorFlowYOLO NanoMATLABSimulinkCWinAPIRS-422MIL-STD-1553Sensor FusionEdge MLReal-Time Systems