Everyday mapping : the design, construction and calibration of a portable surveying device

• Main Author: Scott, Terry
• Other Authors: Newman, Paul
• Published: University of Oxford 2017
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.736061

The contributions of this thesis can be categorised into two main areas. (1) The design, construction and testing of a portable mapping device, and (2) the automatic cross-calibration of cameras with laser scanners. The mapping device, whose entire chassis is manufactured using a 3D printer, is designed to produce high quality 3D maps using a suite of cameras and laser scanners. A key feature of the unit is that it allows for multiple sensors configurations, thus enabling it as a versatile research tool. At the time of writing, the model produced for this thesis was in regular use by the Oxford Robotics Institute and a commercial company. The second component of this thesis is partly an outgrowth of (1); replacing or repositioning sensors on the device requires recalibration of those sensors. We propose and provide analysis for two automatic methods of calibration which use observations of the robot's surroundings to register a camera and laser. The first method uses the information theoretic measure normalised information distance, which is found to perform well but only under certain scene conditions. This leads to our proposal of a scene selection policy which ensures only effective scenes are used for the registration. The second camera - laser calibration method relies on the automatic extraction and matching of image point features between modalities. Lastly, we propose a unifying calibration framework that formulates multi- ple sensor registrations as a pose graph constrained by known inter-sensor measurements. Using an alternating optimisation we are able to treat each expensive registration (as part of a global optimisation) independently. Calibration methods are tested in diverse environments of on-road driving, warehouse and urban pedestrian sidewalks. We demonstrate successful, accurate calibration in each case.