case://study-004
Computer VisionProject 004
SkyRescue Vision
SkyRescue Vision is a software contribution to a UAV-SAR pipeline where I worked on improving human detection and tracking performance for real-time RGB stream analysis.
+25%
Tracking
Real-time
Pipeline
MAVLink
Workflow
Computer Vision
Category
The Problem
What Wasn't Working
Search-and-rescue video feeds are noisy and fast-moving, which makes stable tracking difficult and increases the chance of missed detections during critical operations.
The Solution
How I Fixed It
I contributed to the RGB-stream detection and tracking workflow, improving multi-object tracking stability by tuning SORT association with Kalman-based adjustments for more reliable target continuity.
Stack
Technologies Used
Python
YOLO
OpenCV
SORT
Kalman Filtering
Results
Key Outcomes
✓
Real-time UAV-SAR human detection pipeline support✓
Improved tracking stability by about 25%✓
Integrated YOLO, OpenCV, and SORT workflow✓
MAVLink-ready deployment direction