A.M. Andreev(1), V.S. Kolesnikov(2), V.M. Chernenkiy(3)

1,3 Bauman Moscow State Technical University (Moscow, Russia)

2 Penza State University (Penza, Russia)

1 arkandreev@gmail.com; 3 iu5vmch@mail.ru

Formulation of the problem. Today, robotics is growing more rapidly than ever thanks in large part to AI. Computing power is growing, and this opens up opportunities for applying AI algorithms to low-power devices. The difficulty lies only in testing. If, in the case of regression or classification models, you can take a ready-made dataset, build a model and evaluate the learning outcomes, then in the case of robotics, you need a tangible prototype, on which you can directly test the operation of both the algorithm itself and the system as a whole. There are few such solutions on the market or they are expensive. This article solves the problem of building such a prototype on which it is possible to test computer vision algorithms.

Goal. Design a platform for a ground wheeled robot for testing computer vision algorithms.

Results. The designed prototype is divided into 3 main modules - a power module, a driving module and a main computing module. Each module is independent enough from the others to be modified individually. The main devices of the developed prototype are Arduino and Raspberry pi 4. Raspberry pi 4 is a device that processes computer vision algorithms. The Raspberry pi 4 runs on an ARM v8 processor. For testing the prototype, an algorithm for drawing a trajectory of movement, written specifically for this processor architecture, based on the calculation of the optical flow using the Kanade-Lucas-Tomasi method - ARM-VO, is suitable.

Practical significance. This prototype has a relatively low cost, and therefore is widely available for use. It will be useful to developers of mobile ground-based unmanned robots and developers of computer vision systems both as an inexpensive tool for studying computer vision and robotics, and for testing the algorithms being developed. The prototype allows testing computer vision algorithms as part of a robotic system. The developed device can be easily modified, which makes it possible to modify it for various purposes.

A.M. Andreev¹, V.S. Kolesnikov², V.M. Chernenkiy³

1,3 Bauman Moscow State Technical University (Moscow, Russia)

2 Penza State University (Penza, Russia)

1 arkandreev@gmail.com; 3 iu5vmch@mail.ru

1. «Treking tochek. Lucas-Kanade» [Jelektronnyj resurs]. Data poslednego obnovlenija: 10.09.2011. Data obrashhenija: 20.05.2020. URL:http://blog.scaytrase.ru/computer_vision/369. (In Russian).
2. «ARM-VO: 8 FPS monocular visual odometry on Raspberry Pi 3» [Jelektronnyj resurs]. Data obrashhenija: 20.05.2020. URL:http://imrid.net/?p=4045; (In Russian).
3. «ARM-VO: an efficient monocular visual odometry for ground vehicles on ARM CPUs» [Jelektronnyj resurs]. Data poslednego obnovlenija: 31.05.2019. Data obrashhenija: 20.05.2020. URL:https://link.springer.com/article/10.1007/s00138-019-01037-5. (In Russian).
4. «Raspberry Pi 4 vs Raspberry Pi 3B+: Battle of the Pis 2020» [Jelektronnyj resurs]. Data poslednego obnovlenija: 08.02.2020. Data obrashhenija: 22.05.2020. URL:https://www.youtube.com/watch-v=D_sXMSe1bpo; (In Russian).
5. «Arduino Blue» [Jelektronnyj resurs]. Data obrashhenija: 20.04.2020. URL: https://sites.google.com/stonybrook.edu/arduinoble. (In Russian).