Problem
A human-following robot has to keep a stable target under exactly the conditions that make vision difficult: occlusion, lighting shifts, pose changes, and moving camera perspective.
Back to systemsDesigned for illumination, occlusion, pose, and camera-motion challenges from the start. Connected perception output to motor control and path planning. Validated behavior through simulation and robotics integration.
National Centre of Robotics & Automation
Human-Following Robot
Research Internee / Sep 2020 - Jun 2021
Built human-tracking robotics logic across illumination changes, pose variation, occlusion, and camera motion using visual features, K-D tree classification, Kalman filtering, ROS 1, Gazebo, and Dijkstra planning.
Kalman trackingROS 1Gazebo
My Role
Developed the vision tracking and robotics workflow that connected visual target association, state estimation, simulation, motor control, and navigation.
Impact
A robotics prototype that combined computer vision tracking, state estimation, simulation, and mobile navigation into one working perception-to-motion chain.
System Architecture
- Camera tracking extracts point-based visual features.
- K-D tree classification supports target association.
- Kalman filtering stabilizes target state over time.
- ROS 1, motor control, Gazebo simulation, and Dijkstra planning connect perception to motion.
Tools
PythonROS 1GazeboKalman filterDijkstra
Implementation Highlights
The useful work sits between camera signal and production decision.
Next Step