Catadioptric mobile robot image mosaicing for vehicle undercarriages

Submission note: A thesis submitted in total fulfilment of the requirements for the degree of Doctor of Philosophy to the Department of Electronic Engineering, School of Engineering and Mathematical Sciences, Faculty of Science, Technology and Engineering, La Trobe University, Melbourne.

This thesis investigates and proposes algorithms and techniques to solve the problem of safely imaging the undercarriage of a stationary vehicle to identify explosives or other contraband items. Most current solutions involve humans working very close to vehicles (mirror mounted on a pole), a moving target vehicle (driving over road embedded cameras) or require a high degree of user control (most tele-operated mobile robots). The investigation and work proposed in this thesis uses a mobile robot fitted with a wide angle catadioptric camera system. The catadioptric camera is used to build up a wide-angle image mosaic to provide operators with a view of the easily accessible areas of the vehicle undercarriage. Five main research contributions are described within this thesis: Firstly, a novel binary search pose estimation algorithm is used to identify the pose of the mobile robot with respect to a target vehicle. This is based on the identified locations of each of the target vehicle's tyres and hence allows the robot to plan a path to scan the vehicle. Secondly, an image recognition algorithm is used to identify each of the tyres for a target vehicle from a still photograph and hence assign location-based parameters to each of these tyres for use in the pose estimation algorithm. Thirdly, the use of a catadioptric (wide-angle mirror-based) camera allows larger areas of the target vehicle to be scanned with less scanning traversals. Fourthly, modified optical mouse sensors are characterised and studied as a kinematic- independent solution to augment traditional wheel odometry which is plagued by issues such as wheel slippage. Algorithms to correct for velocity dependency and height induced errors are described and evaluated. Finally, an odometry coupled system is described and evaluated to both significantly improve computational performance for template matching but also to create a rotationally invariant template matching algorithm with which to generate the vehicle mosaic. The contributions are evaluated through simulation, on robotic gantry hardware and through the implementation of a mobile robot which performed imaging underneath both calibration patterns and real target vehicles with simulated contraband devices axed to the undercarriage.