Computer Vision Engineering
Building robust perception systems that enable machines to see, understand, and interact with the physical world in real-time.
Computer Vision Toolkit
Object Detection
I implemented YOLO architectures and fine-tuned models to locate and classify entities in dynamic environments. I achieved high-accuracy and low-latency detection by optimizing the network for our specific use cases.
Image Segmentation
I developed segmentation models to generate pixel-perfect boundaries, giving our robots precise spatial awareness. I used this to successfully identify drivable areas, detect obstacles, and create accurate depth maps for safe navigation.
DEEPSORT Tracking
I integrated DEEPSORT to enable persistent identity tracking across complex video frames. I ensured our system could reliably maintain object states through occlusions and assign unique tracking IDs in real-world scenarios.
Multi-Camera Systems
I built sensor fusion pipelines that integrated LIDAR and Camera data. I managed the extrinsic calibration, synchronized multiple sensor streams, and successfully mapped 2D RGB data onto rich 3D point clouds.
Edge Deployment
I deployed our real-time computer vision models directly to edge devices like the Jetson Nano and Raspberry Pi. I leveraged GPUs, TensorRT, and CUDA to achieve maximum hardware acceleration and significantly boost our FPS.
Dronaid
AI Systems Engineer | Manipal, India
Jul 2022 - Sep 2023
Computer Vision on Edge
Integrated CV models across hardware, embedded & software layers. Improved YOLO realtime inference from 5 to 12 FPS by switching from Python to C++, tuning precision, and applying GPU optimizations (Triton, TensorRT, NSight, CUDA).
Embedded Systems
Worked with Jetson Nano and Raspberry Pi to build software & CV for drones. Integrated camera & LIDAR to CV models via ROS. Deep dived into Linux camera pipelines.
CV Perception Pipeline
YOLO for object detection -> Image segmentation for precise boundaries -> DEEPSORT for tracking -> Mapping coordinates with GPS & LIDAR.
Memory-Compute Profiling
System-level profiling to identify CPU, GPU, Memory, I/O bottlenecks. Troubleshooted end-to-end issues across all layers.
C++ Parallel Programming
Designed multi-threaded processing to separate capture, preprocessing, and inference for stable frame rates using OpenMP & MPI.
Exposure
Collaborated with 40+ members, gaining exposure to microcontrollers, networking, electronic subsystems, and manufacturing.
