AI drone software for intelligent operations.

Examines differences in class detection and localisation across commonly used open-source vision models.

The comparison highlights why dataset quality, model selection and domain-specific tuning are essential for reliable deployment.

Combines complementary detection and tracking methods to maintain target identity through motion and changing viewpoints.

The approach improves continuity and resilience when individual models encounter uncertainty or partial obstruction.

Demonstrates target localisation across thermal imagery where colour and conventional visual detail are limited.

Persistent bounding-box tracking supports navigation and monitoring in low-light, obscured and temperature-contrast environments.

Evaluates tracker behaviour using visual samples and trajectory comparisons across thermal and RGB datasets.

The analysis helps identify drift, missed detections and the operating conditions that affect target-lock stability.

Shows how a selected high-priority object can remain the primary track while other vehicles stay visible in the scene.

This supports focused surveillance workflows where operators must preserve identity on a mission-relevant target.

Illustrates operator-directed selection of different targets within the same operational video feed.

The system can follow the commanded vehicle while retaining contextual awareness of other objects and scene activity.