Object Detection
& Tracking

Overview

ASTROC's AI Vision pipeline enables real-time detection, classification, and tracking of 90+ object classes directly from drone-mounted cameras. The system is optimised for aerial perspectives, handling challenges like small object size, occlusion, and variable altitude.

Technical Stack

The detection pipeline is built on ASTROC Vision System (nano and medium variants) optimised with ASTROC Nebulae for ASTROC Edge deployment. The system processes 1080p video at 15-25 FPS with sub-100ms latency, enabling real-time overlay of bounding boxes and class labels on the video feed.

Custom Training

Beyond the standard detection classes, ASTROC has custom-trained models for domain-specific detection: vehicle types (auto, truck, bus), agricultural objects (crop health indicators), and security-specific items. Transfer learning from large datasets ensures high accuracy even with limited domain-specific training data.

Integration

The detection results are streamed to the ASTROC Sentinel GCS in real-time, where operators can view detected objects overlaid on the map view. The system supports configurable confidence thresholds, class filtering, and alert triggers for specific detection events.