First of all, thank you for the good research paper and presentation contents. I would like to know more about the research environment used in this study, and also in detail about the equipment & operating systems, development languages and software libraries you used.
Thank you very much for your interest in this article. This experiment uses the YOLOv3-tiny and KCF models, while the NX development board is the core of the whole embedded terminal deployment. By connecting a camera to capture image information and using TensorRT to accelerate the YOLOv3-tiny model, the detection model is ported to the NX development board and the red ball positioning and size information is output to the LCD through target detection. NX relies on the UART serial port to communicate with the car's motion module to control the car's left/right/back/forward motion. In addition to this, NX requires the installation of common libraries such as OpenCV and JetPack. In this paper, the YOLOv3-tiny target detection model was built using PyTorch, a deep learning framework that supports parallel training on multiple cards, and the CUDA library was used to speed up the computation on the graphics card. The experiment is hosted on Ubuntu 18.04.5 LTS with an Intel Xeon Gold 6149 processor and a NVIDIA RTX 2080Ti graphics card to meet the model training requirements. In order to run the target detection model on the NX platform, the trained model needs to be accelerated by TensorRT. The experiments were conducted using a server to train a PKUAIP dataset.
