Real-Time Thermal Obstacle Detection

Abstract

Detecting obstacles during low-altitude operations presents a critical challenge for mobile platforms navigating complex environments. This project examines a sophisticated thermal obstacle detection system designed to identify vertical structures—such as power lines, poles, and towers—using thermal infrared imagery.

The Challenge

Low-altitude operations face a persistent challenge: detecting vertical obstacles in the navigation path. Traditional systems using visible-spectrum cameras suffer from:

Poor performance in low light

Inability to operate at night

Difficulty distinguishing obstacles from complex backgrounds

Thermal infrared imaging presents a compelling alternative, allowing operation in complete darkness with superior contrast for man-made structures.

System Architecture

Thermal Image Preprocessing

Defective Pixel Correction

Bilateral Filtering for edge-preserving smoothing

Roll Correction using IMU data

CLAHE Enhancement for local contrast

The Big-Small Map Algorithm

A classical computer vision technique that exploits vertical obstacles' characteristic thermal signatures:

The algorithm detects intensity transitions by comparing patterns between adjacent columns, generating a score map where high values indicate likely obstacle locations.

Deep Learning Classification

ResNet50 architecture performs binary classification on candidate patches, distinguishing true obstacles from false positives with hierarchical feature learning.

Temporal Intelligence

Multi-frame tracking maintains obstacle candidates across frames, requiring persistence before confirmation to dramatically reduce false positives.

Optical Flow Integration

Lucas-Kanade pyramidal tracking maintains continuous obstacle tracking between deep network inference frames.

Key Innovations

Hybrid Architecture: Combines efficient classical methods with powerful deep learning

Real-Time Performance: Meets strict latency requirements for autonomous navigation

Robust Tracking: Multi-frame consistency filtering eliminates transient false positives

Thermal-Specific Processing: Exploits unique properties of thermal imagery

Impact

The system enables safe autonomous navigation in challenging environments where traditional obstacle detection fails, with applications in aviation, robotics, and autonomous vehicles.