Year: 2019
Type: dissertation
Abstract: Odometry is the problem of estimating the motion of a moving platform relative to its \nenvironment without measurements to a fixed reference point. This is a critical problem \nfor mobile robotics applications where measurements to a fixed reference point are not \nalways assured, such as self-driving cars operating in GPS-denied environments and with \nphysical landmarks potentially obscured by other traffic. Even when a fixed reference is \navailable, motion estimates still improve position estimates by constraining the change in \nposition over short timescales. \nA fundamental limitation of odometry is that assuming some non-zero error, the path \nproduced by integrating odometry will always diverge from the true path. The rate of \ndivergence depends on the ego-motion and the environment. Aggressive motions such \nas rapid rotation or high acceleration are likely to cause greater error because they are \nmore difficult to model compared to more sedate motions. Odometry that functions by \ncomparing consecutive sensor samples to determine motion can exhibit greater drift due \nto high velocity because high velocity reduces the overlap between sensor samples. Lastly, \nunstructured portions of the scene may not contain useful information to fully constrain \nthe ego-motion. A good example is moving next to a flat featureless wall since observations \nof that wall only constrain perpendicular motion. \nIn mobile robotics, both camera and lidar are commonly used for odometry. Lidar is a \nsensor technology that measures the time of flight of laser pulses to collect range-bearing \nsamples from the scene. Lidars are used instead of cameras for certain applications despite \ntheir relatively high cost because lidars are not affected by ambient lighting conditions; \n \nthey do not suffer from glare, or have to trade-off motion blur and sensitivity in low- \nlight conditions. An ancillary benefit of using lidar is that the direct range measurement \n \ncapability of lidar removes complexity from the odometry algorithm because the distance \nof sampled points does not have to be estimated from multiple sensor measurements. \nLidar odometry algorithms already exist yet there are opportunities for improvement. \n \nThe intensity information collected by lidar commonly goes unused, with few of the top- \nperforming lidar odometry algorithms on the Kitti odometry dataset leveraging it. As- \nsuming that scenes lacking both geometric and appearance information are less likely than \n \nthose lacking only geometric information, then a lidar odometry algorithm that leverages \nboth types of information will be more robust than an algorithm that relies only on one \ntype, all other things being equal. Robustness is desirable because it makes performance \npredictable. \nThe main contribution in this thesis is a lidar odometry algorithm that uses both \nappearance-based intensity landmarks as well as geometric landmarks to model the scene. \n \nThe intensity landmarks are gradient edges that model features in the environment such as \nlane markings while the geometric features are geometric edges and planes. Feature points \nare selected from multiple lidar scans and matched to the landmarks. The landmarks \nand the trajectory are jointly optimized within a sliding window filter. The addition of \nintensity landmarks improves performance in most cases compared to using geometric only \nlandmarks. \nEvaluation is carried out using the Kitti odometry metric on the Kitti odometry dataset \nas well as a dataset collected in Waterloo Ontario. Average drift on the Kitti training \nset was 4.6-9.6 deg/km and 0.71%-2.8% translation drift. Average drift on the Waterloo \n \ndataset was 3.1-3.4 deg/km and 1.1%-1.2% translation drift. Compared to a baseline con- \nfiguration using only geometric landmarks, adding appearance-based landmarks produced \n \na slight performance improvement on average across the Kitti subsequences as well as the \nWaterloo dataset, with up to a 13% reduction on translation error and a 6% reduction \n \nof rotation error for specific subsequences. However, there are also subsequences that ex- \nhibited worse performance with the addition of appearance-based features. In conclusion, \nthe evaluation shows that the addition of appearance information can lead to better per- \nformance and that the method presented would benefit from additional work on further \ndeveloping the scene appearance model and how it is applied.
Source: UWSpace (University of Waterloo)
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2938685850