Precision Timing and Measurement for Inference with Laser and Vision
University of Oxford Oxford United Kingdom
Pagination or Media Count:
This thesis is about precise acquisition and inference of 3D range data. The intended application domain is mobile robotics. We investigate a number of key issues in the data gathering process, showing that a careful treatment of each stage in the pipeline is vital for producing dense accurate representations. We describe a low cost 3D laser system capable of generating high quality data from a continuously moving platform. The hardware, data capture, calibration and processing techniques we have developed allow us to produce remarkably detailed point clouds. Our laser systems rapid scanning of the environment enables the correction and augmentation of the robots odometry system, by tracking planar features in consecutive sweeps of the environment. An essential part of the data acquisition pipeline is accurate timestamping of data from different sources. We describe a new and very efficient algorithm for the rapid on line synchronization of computer clocks distributed over a network. The algorithm, known as TICSync, is capable of achieving performance measured in Parts Per Billion and deals naturally with common clock upset events. We also contribute a method for fusing point clouds with camera images to produce dense and accurate range maps of much higher resolution than the input range data. We make use of structural similarities in range and intensity data. Our use of a 2nd-Order smoothness prior allows the method to infer surfaces of arbitrary slope, as well as to reconstruct curved surfaces where appropriate.
- Operations Research
- Optical Detection and Detectors