Iterative Snapping of Odometry Trajectories for Path Identification

Abstract

Abstract. An increasing number of mobile devices are capable of automatically sensing and recording rich information about the surrounding environment. Spatial locations of such data can help to better learn about the environment. In this work, we address the problem of identifying the locations visited by a mobile device as it moves within an indoor environment. We focus on devices equipped with odometry sensors that capture changes in motion. Odometry suffers from cumulative errors of dead reckoning but it captures the relative shape of the traversed path well. Our approach will correct such errors by matching the shape of the trajectory from odometry to traversable paths of a known map. Our algorithm is inspired by prior vehicular GPS map matching techniques that snap global GPS measurements to known roads. We similarly wish to snap the trajectory from odometry to known hallways. Several modifications are required to ensure these techniques are robust when given relative measurements from odometry. If we assume an office-like environment with only straight hallways, then a significant rotation indicates a transition to another hallway. As a result, we partition the trajectory into line segments based on significant turns. Each trajectory segment is snapped to a corresponding hallway that best maintains the shape of the original trajectory. These snapping decisions are made based on the similarity of the two curves as well as the rotation to transition between hallways. We will show robustness under different types of noise in complex environments and the ability to propose coarse sensor noise errors.

Topometric localization on a road network

“…Localization is achieved by matching observed trajectories, constructed from visual odometry [9], [12] or IMU-based odometry [22], with a prior map using various techniques, such as Chamfer matching [12], curve similarity matching [22], or street segment matching [9]. Such approaches, however, are subject to drift in situations where the road geometry offers no constraints (e.g., long, straight road segments), particularly when visual odometry is used.…”

Indoor trajectory identification: Snapping with uncertainty

“…Specifically, we build on prior work in [1], [2] and implement a "snapping" algorithm to reconstruct human paths traversed in a real indoor environment, given an existing map of that environment. Our algorithm searches for all plausible paths within specified error bounds using the map and a segmented trajectory derived from accelerometer and gyroscope measurements.…”

“…It was first used to handle indoor path identification tasks with a wheeled robot [1], [2], and proved robust in several real settings. Previous efforts relying only on cell phone odometry used probabilistic techniques like particle filters [7], which have strong independence assumptions between measurements.…”

“…In this paper, the "snapping" technique of [1] is applied to humans for the first time, and the only sensors used are the accelerometer and gyroscope found in a Samsung Galaxy S4. The process of deriving a walker's trajectory from raw data is as follows:…”

O-Snap: Optimal snapping of odometry trajectories for route identification

“…We aim to overcome drift errors by reshaping the odometry trajectory to fit the constraints of a given topological map and sensor noise model. Prior works use iterative search algorithms that are susceptible to local maximas [15], which means that they can be misled when faced with ambiguous decisions. In contrast, our algorithm is able to find the set of all routes within the given constraints.…”