Outdoor Radar Mapping Using Measurement Likelihood Estimation

This paper fuses target detection and occupancy mapping theory to develop an improved method for outdoor mapping with a radar sensor. It is shown that the occupancy mapping problem is directly coupled with the signal detection processing which occurs in a range sensor, and that the required measurement likelihoods are those commonly encountered in both the target detection and data association hypotheses decisions. The classical binary Bayes filter approach generally treats these measurement likelihoods as fully known deterministic values. From examination of radar detection theory it is shown these likelihoods are only deterministic under a number of unrealistic assumptions which make it impractical for a real radar system used in a outdoor environment. An algorithm is therefore presented which jointly estimates the measurement likelihoods of each target in the environment and uses a particle filter to propagate their corresponding occupancy estimates. The ideas presented in this paper are demonstrated in the field robotics domain using a millimeter wave radar sensor, and comparisons with laser based maps as well as previous radar models are shown.