Methodology of Evaluation of Low Cost Electronic Devices : Raspberry PI and Nvidia Jetson Nano for Perception System Implementation in Robotic Applications

In order to realize perception systems on low cost embedded electronic devices as Raspberry PI 3, 4, and Nvidia Jetson Nano, an evaluation methodology to test these devices is proposed. These systems are dedicated for robotics applications for consummer uses as domestic autonomous robots, and systems to help visually impaired people to move safely. The methodology consists in choosing representative low level vision algorithms and in evaluating them using various implementations. CPU and GPU implementations are evaluated. The interest of the methodology is the speed of the evaluation. Indeed, the implementation of the reference algorithms is much faster than the implementation of the algorithms of the targeted application. The coarse time scale, between the treatments chosen as a reference and those to be implemented for the targeted application, gives an idea of the possibility of use. This methodology is then performed to evaluate and to compare the three devices : Raspberry PI 3, 4, and Nvidia Jetson Nano. The results shows the interest of the use of the GPU. The easy access to the GPU, and the simplicity of programming in Cuda is a great advantage for the Jetson Nano.