Dynamic Target Prediction Based Obstacle Avoidance for Inspection Robot in Real-Time Hazard-Trending Path Planning

In fire hazard environments, the inspection robot performs for early fire detection with real-time obstacle avoidance. However, the dynamic target usually blocks the fire hazard when the robot moves closer. Addressing this challenge, a real-time hazard-trending path optimization strategy based on dynamic target trajectory prediction is developed for avoiding dynamic target blocking. Firstly, the limited field of view of the thermal imaging sensor leads to detection blind spots. Thermal enhanced dynamic window-based planner is proposed by integrating temperature information into local path planning. Secondly, prediction-aware NavFn is developed for robotic path replanning. Combining YOLOv7 and DeepSORT algorithm, the dynamic target is recognized and tracked. Linear extrapolation is employed to predict trajectory. Through dynamically replanning the global path to avoid potential collision points, the robot avoids the dynamic target in advance. Finally, experiment results are analyzed by comparing with methods regarding avoidance success rate, path deviation, and hazard missing rate.