The efficient route learning using image processing

Abstract

With the proliferation of LBSs (Location-Based Services) using information on user locations, a significant amount of research on location models from GPS data has been actively conducted. Existing methods have trouble with high complexity as they calculate for similarities using a combination of spatial and temporal GPS data sequences, which often deteriorates accuracy. To reduce complexity and enhance accuracy, we have adapted image processing algorithms instead of mathematical/probabilistic computations. First, we consider each trip of the user as an image consisting of lines on a grid, as opposed to a GPS data sequence. By comparing the images, we extract lines with similar patterns using a thinning algorithm. Through this process, we decrease the complexity involved in learning the spatial patterns of personal routes. Next, we add temporal information into the learned lines and determine personal routes using the similarity function of temporal and spatial information. Our contributions to learning personal routes are in terms of reducing complexity and enhancing performance, such as accuracy and speed, by restating the computational problem as an image processing problem and narrowing down the problem space. Experiments show recall and precision rates of around 90% and 96%, respectively. © 2012 American Scientific Publishers All rights reserved.