Mobile robot sonar interpretation algorithm for distinguishing trees from poles

We describe an algorithm to automatically distinguish trees from round metal poles with a sonar system packaged on a mobile robot. A polynomial interpolation of the square root of the backscattered signal energy versus scan angle is first plotted. Asymmetry and fitting error are then extracted for each sweep across the object, giving a single point in an abstract phase space. Round metal poles are nearer to the origin than are trees, which scatter the sonar more irregularly due to lobulations and/or surface roughness. Results are shown for 20 trees and 10 metal poles scanned on our campus.