Abstract: Due to the rapid development of affordable, intelligent, and mobile devices, especially in the area of wearable computing (e.g., Google Glass, etc.), the implementation of applications for computational vision is more and more feasible.
In this context, we propose a new concept for a pedometer, which is based on visual feature points.
The introduced method is part of a visual odometry procedure, which is used for positioning on basis of optical feature points and their corresponding vectors.
In order to realize the pedometer, video recordings shot from a first person perspective become analyzed. In a first step, feature points are extracted out of each of the video’s frames. Subsequently, a pace movement is calculated by using the euclidian norm between equal feature points of successive frames.
In the following, the concept is compared to classical pedometers, which are commonly based on an accelerometer or related inertial sensors.
Besides the successful recognition of a user’s paces, we take a brief look at popular techniques for feature point processing, e.g., SURF, ORB, and BRISK, regarding their suitability for visual pace detection, which is also subject of this paper.
Especially their characteristics concerning robustness and computational costs for successful pace recognition are relevant in this context.
Konferenz: 1st GI Expert Talk on Localization