AP P LYING P ERSON LOCAL IZAT ION TO CONTROL THE FARMCARRI ER: A S EMI -AUTONOMOUS VEHICL E

Ven, van der, Wouter

AP P LYING P ERSON LOCAL IZAT ION TO CONTROL THE FARMCARRI ER: A S EMI -AUTONOMOUS VEHICL E

Files

Ven, vd, W. s-4356403-2021.pdf (12.73 MB)

Authors

Ven, van der, Wouter

Issue Date

2021-05-01

Language

en

URI

https://theses.ubn.ru.nl/handle/123456789/16158

Abstract

Farmers have to collect dead or physically injured broilers inside the poultry house, currently a (powered) wheelbarrow is used to move them. The FarmCarrier is developed by LiveStock Robotics to assist poultry farmers with this work. Currently only low-level control with the use of a remote control is available for the FarmCarrier. In the future, the FarmCarrier has to be able to autonomously follow a person through the poultry house. In this research, I present a pipeline to follow the farmer using an RGBD-sensor. Three different networks (MobileNet, YOLO and Tiny-YOLO) and two different embedded hardware systems (Jetson Nano and Jetson TX2) were compared on their processing speed and their localization accuracy in two different scenarios. The first scenario was outside with good lighting and free movement for the FarmCarrier, the second scenario was inside with limited movement and poor lighting. Experiments pointed out that that MobileNet had too few frames in which a person was detected, while YOLO and Tiny-YOLO had comparable results in terms of the amount of detections. The localization accuracy of YOLO and Tiny-YOLO did not differ significantly, while detection times of YOLO were too high on both embedded hardware systems. The hardware implementation with the Jetson TX2 had a sufficient detection speed where the Jetson Nano did not reach sufficient speeds. Therefore, Tiny-YOLO with the Jetson TX2 presented the best trade-off between performance and computation time for an embedded solution in the FarmCarrier.