Body Language
Detection

Overview

ASTROC's Body Language Detection system uses AI-powered human pose estimation to analyse behavioural patterns from aerial drone footage. The system detects individual humans, estimates skeletal keypoints, and classifies body postures and gestures in real-time for crowd monitoring and security applications.

Technical Approach

The system combines a ASTROC Vision Pose Engine model for human detection and keypoint estimation with a custom proprietary classifier for temporal activity recognition. By analysing sequences of skeletal poses across video frames, the system can classify activities like walking, running, standing, sitting, crouching, and flagging aggressive or unusual body language patterns.

Applications

Deployed for large-event crowd monitoring, border surveillance, and critical infrastructure protection. The system integrates with ASTROC Sentinel GCS to provide operators with real-time alerts when specific body language patterns or crowd density thresholds are detected, enabling proactive security response.

Capabilities

The system can simultaneously track up to 50 individuals in a single frame, estimating 17 body keypoints per person at 12+ FPS. Activity classification runs on a sliding window of 30 frames, providing smooth and stable behaviour predictions.