An evaluation of real-time RGB-D visual odometry algorithms on mobile devices

Vincent Angladon (1, 2) , Simone Gasparini (1, 3) , Vincent Charvillat (1, 3) , Tomislav Pribanić (4) , Tomislav Petković (4) , Matea Ðonlić (4) , Benjamin Ahsan (2) , Frédéric Bruel

1 IRIT-REVA - Real Expression Artificial Life
2 Telequid
3 Toulouse INP - Institut National Polytechnique (Toulouse)
4 FER - Faculty of Electrical Engineering and Computing [Zagreb]

We present an evaluation and a comparison of different Visual Odometry algorithms, selected to be tested on a mobile device equipped with a RGB-D camera. The most promising algorithms from the literature are tested on different mobile devices, some equipped with the Structure Sensor. We evaluate the accuracy of each selected algorithm as well as the the memory and CPU consumption, and we show that even without specific optimization, some of them can provide a reliable measure of the device motion in real time.

Mots clés: Visual Odometry; RGB-D Camera; Benchmark; Mobile device

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Vincent Angladon, Simone Gasparini, Vincent Charvillat, Tomislav Pribanić, Tomislav Petković, et al.. An evaluation of real-time RGB-D visual odometry algorithms on mobile devices. Journal of Real-Time Image Processing, 2019, 16 (5), pp.1643-1660. ⟨10.1007/s11554-017-0670-y⟩. ⟨hal-01654706⟩