Sonar based mapping using PHD filter

In this paper, we present a feature based sonar mapping technique for reconstructing structured underwater environments using sonar images given the robot pose. Underwater environmental elements are represented using Hough transform based line segments. These features are filtered using the Probability Hypothesis Density (PHD) filter under a Random Finite Sets (RFS) framework of the environment. The approach handles sensor miss detections and false alarms for reconstruction of a reliable world model. Furthermore, we demonstrate how a weighted accumulation in Hough parameter space can improve line detection using a forward looking sonar sensor. We show that the performance of the system using a simulated environment where the robot is driven in a 2D plane facing the sides of a simulated underwater arena.