Student Bachelor Thesis: Utilizing LiDAR Data for Human Detection in Fire-Restricted Environments Using NorrSpect AI’s Robot Dog

Title:

1. Background & Motivation

Search and rescue (SAR) operations in fire-restricted environments pose significant risks to human responders. The use of autonomous robots equipped with advanced sensors can help mitigate these risks by identifying human presence in low-visibility conditions caused by smoke and fire.

NorrSpect AI’s robot dog, which is already equipped with mobility and perception capabilities, presents an opportunity to integrate LiDAR-based human detection for enhanced search capabilities in emergency scenarios. This project aims to explore the feasibility of using LiDAR technology to detect humans in fire-affected zones, ensuring efficient and autonomous rescue assistance.

2. Problem Statement

The primary challenge in fire-restricted environments is low visibility due to smoke, heat distortions, and potential infrastructure damage. Conventional cameras and thermal imaging alone may be insufficient in certain conditions. LiDAR sensors, which provide a depth-based 3D representation of the environment, could enhance human detection in such scenarios.

This thesis will investigate:

• How well LiDAR-based human detection performs in smoke-filled environments.

• The real-time processing capabilities of LiDAR data for rapid human localization.

• The integration of LiDAR and AI-based object recognition with NorrSpect AI’s robot dog platform.

3. Objectives

1. LiDAR Data Collection & Processing:

   • Capture point cloud data from LiDAR sensors in different environmental conditions (clear, low-visibility, smoke-filled).

   • Preprocess and filter noise from LiDAR scans.

2. Human Detection & Recognition:

   • Develop ML/AI models to classify human shapes based on LiDAR point clouds.

   • Compare traditional rule-based segmentation (silhouettes, body contours) vs. deep learning approaches.

3. Integration with Robot Dog:

   • Implement LiDAR-based navigation for detecting and approaching humans.

   • Optimize real-time data transmission and decision-making for autonomous movement.

4. Validation & Testing:

   • Test in controlled conditions simulating fire-restricted areas.

   • Evaluate accuracy, false-positive rate, and response time.

4. Methodology

1. Literature Review:

   • Research on existing LiDAR-based human detection techniques.

   • Analyze past applications of autonomous robots in SAR operations.

2. Hardware & Software Setup:

   • Utilize a high-resolution LiDAR sensor (e.g., Velodyne, Ouster, or Hokuyo).

   • Develop a ROS-based framework to interface with NorrSpect AI’s robot dog.

   • Implement AI models using TensorFlow/PyTorch for point cloud classification.

3. Implementation & Testing:

   • Develop an algorithm for human recognition from LiDAR scans.

   • Train models using simulated datasets & real-world fire rescue scenarios.

   • Conduct experiments in fire-restricted conditions (smoke-filled rooms, low light, high temperatures).

4. Performance Evaluation:

   • Measure accuracy, speed, and reliability of human detection.

   • Compare against conventional thermal imaging and RGB cameras.

   • Propose improvements and limitations.

5. Expected Outcomes

• A LiDAR-based detection system capable of identifying humans in fire-restricted environments.

• Real-time navigation and human approach algorithms integrated with the robot dog.

• A comparative analysis of LiDAR vs. other detection modalities in low-visibility conditions.

• Open-source dataset and software implementation for future research.

6. Potential Challenges & Risks

• LiDAR performance limitations in extreme heat or dense smoke.

• Computational constraints for real-time processing on the robot.

• Integration complexity with existing hardware/software of NorrSpect AI’s robot dog.

• Limited availability of real-world testing environments.

7. Timeline (6 Months Plan)

MonthTask1Literature review, hardware selection2Data collection, preliminary AI model development3Integration with robot dog (LiDAR-based perception)4Testing in controlled environments Optimization, real-world scenario validation6Thesis documentation & final presentation

8. Resources Required

• NorrSpect AI Robot Dog

• LiDAR Sensor (e.g., Ouster, Velodyne)

• GPU-enabled computing unit for AI processing

• ROS (Robot Operating System)

• Simulation Environments (Gazebo/PyBullet for virtual tests)

• Fire safety-approved testing lab

• 
  

9. Conclusion

This project will advance the use of autonomous robotics and AI-driven LiDAR perception in search-and-rescue missions. By integrating LiDAR-based human detection with NorrSpect AI’s robot dog, the proposed system will improve efficiency in locating humans during fire emergencies, reducing risks for human responders.

Reach Out to - ulrik.s@norrspect.com