ACRONA AR CORE includes the two following approaches of global positioning:
GPS-based: uses incoming GPS and built-in compass data;
MAP-based: uses recognition of spatial anchors which are obtained from the ARCONA server.
The both approaches use binding of the current SLAM trajectory with incoming information represented by the stream of GPS and compass messages for the first approach and recognized spatial anchors for the second one.
Starting with the lowest precision they increase the accuracy while new information comes until the maximum possible precision is reached. Roughly, the necessary conditions to achieve this best precision are:
1) the user should reach a certain distance from its starting point (regardless of its trajectory);
2) a certain number of GPS messages (for GPS-based) or spatial anchor recognitions (for MAP-based) should be collected.
The GPS-based approach is suitable for extent locations without high buildings and other distorting GPS signal structures. Examples of such locations are: parks, gardens, willages, majority of city historical centers and so on. Positioning with this approach has a little computational cost.
The MAP-based approach is intended to use for quite compact locations containing perciptible non-deformable objects like buildings, rocks, monuments that can be recognized by a user device using its on-board camera.
This approach provides greater accuracy. But, to use it, a proper set of spatial ancors should be created for the corresponding location before. Such sets can be created automatically on the server side from the corresponding sampled data packages which are made and sent by a specially designed mobile application.
Each anchor can be recognized within the range up to 1.5 m and angular aperture up to 60 dg. Increasing of these range and aperture is possible but leads to significant precision losses.
Anchors can form some kind of sparse field covering the corresponding location or be placed at the most frequently visited points. To avoid mobile device overloading, all the computations of this approach are performed strictly in two threads and only one of them performs massive numerical operations, so in fact no more than one processor core can be occupied.
The both positioning approaches can be used simultaneously supporting each other
Development of the both approaches had been started in year 2019 and the main indices of the first done version (2020) and the newest one are adduced in the tables below . In these tables there are the following denotations:
linear precision in meters - “Lp”; angular precision in degrees - “Ap”; for adduced indices: median : min/max value; remoteness from the starting point in meters - “R”; number of the collected navigation events (GPS messages or anchor recognitions) - “N”. So, “R = 0, N = 1” corresponds to the start case.
Table 1: precision indices of the GPS-based approach
Table 2: precision indices of the MAP-based approach
In addition. Version 2022 of the MAP-based approach has about 2 times increased probability of anchor recognition with comparison with version 2020.
Our SLAM implementation belong to the class of visual-inertial bundle adjustment solutions; distinct from many similar existing ones ours can perform the full-range nonlinear optimization problem on a quite weak mobile device (Apple 6s) with satisfactory performance (10 – 16 fps) and has up to 30% lesser memory consumption.