Design and Implementation of a Real-Time Visual Tracking System for UAVs Based on PSDK

Area of Science:

• Robotics
• Computer Vision
• Control Systems

Background:

• Resource-constrained edge platforms pose challenges for real-time visual tracking in UAVs, requiring a balance between precision and low latency.
• Existing systems often face latency issues due to middleware layers like Robot Operating System (ROS).

Purpose of the Study:

• To design and implement a real-time visual tracking system for UAVs with high precision and low latency.
• To overcome computational limitations on edge platforms.
• To validate the system's stability and tracking performance in diverse field conditions.

Main Methods:

• Developed a Middleware-Free Architecture utilizing DJI Payload Software Development Kit (PSDK) to reduce serialization overhead and end-to-end latency.
• Proposed a Lightweight Asymmetric De-coupled Visual Servoing (ADVS) strategy, incorporating orthogonal kinematic de-coupling and a non-linear dead-zone with dynamics-aware gain scheduling.
• Implemented a Geometry-Aware Fusion strategy for outlier rejection and a Finite State Machine (FSM) for temporal consistency.

Main Results:

• Achieved a significant reduction in end-to-end latency by over 60%, reaching approximately 30 ms.
• Demonstrated robust target tracking at speeds up to 23 m/s across dynamic maneuvers.
• Verified system stability and tracking capability through extensive field experiments.

Conclusions:

• High-performance edge tracking can be achieved through architectural latency minimization and optimized robust control strategies, not just increased model complexity.
• The proposed system offers an effective solution for real-time visual tracking on resource-limited UAV platforms.
• The Middleware-Free Architecture and ADVS strategy represent significant advancements in UAV visual tracking technology.