Terrain Classification for Autonomous Navigation Using Ladar Sensing

Propose a terrain classification algorithm from ladar data for autonomous navigation. The algorithm using multi-feature which including geometrical feature and color feature different from the classical method that based on single geometric feature. Average normal vector covariance matrix of the point and eigenvalue of the coordinate covariance matrix were extracted first and then the algorithm makes a coordinate calibration between camera and laser scanner. A point will get the RGB feature from image after calibration. A classifier is learned using multi-feature vector by employing an EM-GMM model after marking training data by human beings. The result shows that using multi-feature can get higher classification correctness than using single geometric feature.