Scene Modeling from Motion-Free Radar Sensing

Abstract

Radar can offer remarkable advantages for robotic safeguarding, mapping and navigation because it is not as vulnerable to vacuum, dust, fog, rain, snow and dark. These conditions challenge the performance of sonar, laser and stereo in automating construction, mining, agriculture and planetary-exploration. Sonar cannot operate in vacuum conditions, low visibility and scanning mechanisms limit laser's reliability, and stereo is dependent on scene texture and illumination. Another radar advantage is electronic steering. However, radar has shortcomings such as a large footprint, sidelobes, specularity effects and limited range resolution, all of which result in poor perception models. Some of these problems worsen with the introduction of motion-free scanning, which is important for high reliability, size reduction, and higher frame rate. This thesis presents a short-range radar interpreter that accounts for properties such as sidelobes, varying antenna-radiation patterns and noise distribution. The interpreter uses a three-dimensional evidence-grid representation to accumulate the radar data. A radar interpretation model, which accounts for the antenna radiation properties and the signal and noise distributions, populates the grid with each radar observation. The combination of data from disparate vantage points mitigates the ambiguities introduced by the sidelobes, attenuates the noise, and emphasizes the evidence of objects in the scene. To exhibit the capabilities of the interpretation methodology, the thesis presents the results of simulations and experimental activities. A radar simulator provides data for exploring the different aspects of the interpretation methodology. Two experimental tasks, landscape modeling from multiple viewpoints and vehicle-mounted radar interpretation, show the applicability of the technique for robotic perception tasks.