Development of multi-terrain automatic guided vehicle (AGV) using vision assistive system

Abdul Nazar, Muhsin Kamil (2021) Development of multi-terrain automatic guided vehicle (AGV) using vision assistive system. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Development of multi-terrain automatic guided vehicle (AGV) using vision assistive system.pdf - Submitted Version Download (1MB)
	Text (Full text) Development of multi-terrain automatic guided vehicle (AGV) using vision assistive system.pdf - Submitted Version Restricted to Repository staff only Download (5MB)

Abstract

The multi-terrain automatic guided vehicle (AGV) can be implemented in various categories, such as in production line for automatic delivery and for moving heavy objects from one point to another point. In this project the AGV will implement vision assistive system for detecting object and obstacle. Currently, AGV available on the market have some limitations where it was not designed for obstacle detection, obstacle avoidance, moving on different terrains and slope climbing. In this project, the automatic guided vehicle (AGV) prototype has already been fabricated (continuing from previous FYP project). The first objective is to develop algorithm for object detection and avoidance using pretrained YOLO object recognition model. The second objective is to study the effects when using the algorithm developed to menuever an AGV. YOLO pretrained model will be used in the vision system of this project. For this project, python programming language is selected because it is appropriate for novices and it is not difficult to learn. Python programming language is also widely used in vision system because it contains the OpenCV library used for computer vision. Xiaovv 1080P USB Webcam is used to capture the images of object and laptop is used to carry out recognition. Arduino is used to control the motor on the AGV for meneuvering and obstacle avoidance. There will be 4 experiments done in this project. First experiment would be to compare the performance of YOLOV3 model and YOLOV3-TINY model in terms of frame rate. YOLOV3 have frame rate of 18.3 FPS while YOLOV3-TINY have a frame rate of 49.1 FPS. For the second experiment, the objective would be to compare the performance of YOLOV3 model and YOLOV3- TINY model in terms of accuracy. YOLOV3 have accuracy of 66.66% while YOLOV3-TINY have 65.31% accuracy. In the third experiment, performance of the vision system with different confidence threshold values would be tested. 60% confidence threshold was choosen because it has the highest accuracy of 70.36%. Lastly, for the fourth experiment, the effect of delay in the system would be analysed. All the experiment will be repeated 6 times and the result would be observed. In this project, the AGV is expected to manoeuvre without any control by humans.

Actions (login required)

View Item