Research on Path Planning Algorithm Based on Fast Target Detection

Abstract

As a key technology of robot navigation, path planning has garnered widespread attention and has been utilized in various applications such as mobile robots, unmanned aerial vehicles, and human-computer interaction. Recently, several studies advocate constructing semantic maps for path planning in the laboratory stage. However, these approaches require large storage space and high computing resource consumption, making it difficult to meet real-time requirements. We tackle this issue by building real-time semantic navigation map and propose a real-time path planning algorithm based on fast target detection. Specially, we first construct two-dimensional grid map using the Gamapping method and locate the target objection utilizing the object detection algorithm YOLOv3 retrained in an indoor experimental environment. Furthermore, by incorporating the category information and position information of the detected object into the two-dimensional grid map through a coordinate mapping mechanism, we combine the geometric metric information and visual detection information to build semantic navigation map for automatically planning a reasonable path. The experiments conducted on both qualitative and quantitative levels have demonstrated that our method achieves superior performance and practical application value.