Document Type
Honors Thesis
Abstract
Darknet YOLO is a real-time object detection system that is used in this project. YOLO, which stands for You Only Look Once, uses a single convolutional neural network to detect objects in an image. There are various versions of YOLO, all with their own advantages and disadvantages. For the purpose of this project we will be using Tiny YOLOv2, which is a version of YOLO that is lightweight and performs less calculation, giving us a lower accuracy but higher frame rate. Tiny YOLOv2 applies a single neural network to the full image and divides the image into regions and predicts bounding boxes and probabilities for each region. As part of the Senior Design project our team has trained a seven-class object detection model. The classes that can be detected by this model are people, forklifts, trucks, boxes, pallets, pallet jacks and industrial carts. The goal of this Honors thesis was to run this object detection model on different hardware systems to decide the best option. This model is tested on five different systems: Raspberry Pi 3, Asus Intel i5 4th Gen CPU, Nvidia Jetson TX1, Nvidia Jetson TX2 and Nvidia GeForce GTX 980 Ti. After taking into account the frame rate, accuracy and cost of each of these systems, our recommendation is to use Nvidia Jetson TX2.
Publication Date
5-1-2018
Language
English
License
This work is licensed under a Creative Commons Attribution-NonCommercial-Share Alike 4.0 International License.
Recommended Citation
Sardesai, Tanmay, "DETERMINING HARDWARE SETUP FOR TRAINING AND TESTING AN OBJECT DETECTION MODEL FOR USE IN AN INDUSTRIAL SETTING" (2018). 2018 Spring Honors Capstone Projects. 29.
https://mavmatrix.uta.edu/honors_spring2018/29