Environmental Perception and Sensor Data Fusion for Unmanned Ground Vehicle

Abstract

Unmanned Ground Vehicles (UGVs) that can drive autonomously in cross-country environment have received a good deal of attention in recent years. They must have the ability to determine whether the current terrain is traversable or not by using onboard sensors. This paper explores new methods related to environment perception based on computer image processing, pattern recognition, multisensors data fusion, and multidisciplinary theory. Kalman filter is used for low-level fusion of physical level, thus using the D-S evidence theory for high-level data fusion. Probability Test and Gaussian Mixture Model are proposed to obtain the traversable region in the forward-facing camera view for UGV. One feature set including color and texture information is extracted from areas of interest and combined with a classifier approach to resolve two types of terrain (traversable or not). Also, three-dimension data are employed; the feature set contains components such as distance contrast of three-dimension data, edge chain-code curvature of camera image, and covariance matrix based on the principal component method. This paper puts forward one new method that is suitable for distributing basic probability assignment (BPA), based on which D-S theory of evidence is employed to integrate sensors information and recognize the obstacle. The subordination obtained by using the fuzzy interpolation is applied to calculate the basic probability assignment. It is supposed that the subordination is equal to correlation coefficient in the formula. More accurate results of object identification are achieved by using the D-S theory of evidence. Control on motion behavior or autonomous navigation for UGV is based on the method, which is necessary for UGV high speed driving in cross-country environment. The experiment results have demonstrated the viability of the new method.