Probabilistic estimation of unmarked roads using radar

This paper presents a probabilistic framework for unmarked roads estimation using radar sensors. The algorithm models the sensor likelihood function as a Gaussian mixture model. This sensor likelihood is used in a Bayesian approach to estimate the road edges probability distribution. A particle filter is used as the fusion mechanism to obtain posterior estimates of the road's parameters. The main applications of the approach presented are autonomous navigation and driver assistance. The use of radar permits the system to work even under difficult environmental conditions. Experimental results with data acquired in a mine environment are presented. By using a GPS mounted on the test vehicle, the algorithm outcome is registered with a satellite image of the experimental place. The registration allows to perform a qualitative analysis of the algorithm results. The results show the effectiveness of the algorithm presented.