UAV Indoor Exploration for Fire-Target Detection and Extinguishing

Halil Utku Unlu, Dimitris Chaikalis, Athanasios Tsoukalas, Anthony Tzes

Abstract: A UAV equipped with an RGB-D camera and a downward facing optical flow sensor is set to explore an unknown indoor 3D-volume. The UAV’s visual odometry is accomplished through the fusion of IMU, depth readings and the optical flow sensor. While the UAV is moving using a skeletal path to avoid collisions between the surrounding obstacles, it computes features from the environment. These features assist the UAV to localize itself in a self-identified map using the RTAB-Map SLAM. The UAV moves to the fire-alike target-set and for the unexplored area it uses a Chebyshev shortest path from the boundary free cells to the boundary of the target-set. Due to SLAM-induced errors, the UAV employs virtual potential fields and local depth measurements for obstacle avoidance. Upon detection of the target, comprised as a fiducial marker to emulate a fire-source, the UAV uses a ballistic trajectory to propel its payload (fire extinguishing material) towards the target. Simulation studies using Unity for photo-realistic indoor imaging, Gazebo for the UAV dynamics and ROS as intermediate between these components is used to validate the suggested scheme.