AI-powered autonomous drone navigation, visual target identification, click-to-lock tracking, and decoupled asynchronous flight control.
Aervion is an applied computer vision and drone flight control engineering project built to explore autonomous vision-guided following on lightweight quadcopter platforms (DJI Tello SDK).
The system integrates live video feed streaming, real-time face detection, target tracking, click-to-lock target selection, voice control triggers, and an interactive browser-based dashboard built with Streamlit.
In initial prototypes, running deep neural face detection on the main loop caused severe frame-rate drops. Whenever image preprocessing or inference took longer than 80ms, the flight-control loop missed vital command cycles.
Quadcopter flight stability relies on a continuous high-frequency heartbeat and Proportional-Integral-Derivative (PID) correction loops. Coupling expensive image inference directly to the flight thread led to control latency, erratic yaw/pitch corrections, and safety disconnects.
The primary engineering objective was to achieve smooth, jitter-free target following without sacrificing flight safety or control frequency.
We redesigned the system into an asynchronous, multi-threaded pipeline that isolates heavy vision processing from flight telemetry:
Result: Even if image inference fluctuates due to lighting conditions or complex scenes, the flight stabilization thread continues uninterrupted, ensuring smooth, predictable quadcopter tracking.