Fast trajectory matching using small binary images

Abstract

This paper proposes a new trajectory matching method using logic operations on binary images. By using small binary images we are able to effectively utilize the large word size offered in modern CPU architectures, resulting in a very efficient evaluation of similarities between trajectories. The efficiency is caused by the fact that all bits in the same word are processed in parallel. Representing trajectories as small binary images has other advantages, such as a low space requirement and good noise resistance. The proposed method is evaluated on a publicly available dataset, and is compared to the more sophisticated Longest Common Subsequence (LCSS) method. In addition, synthetic experiments show the good efficiency and accuracy of the proposed method, enabling real time trajectory retrieval on databases with millions of trajectories.