Object Detection for Unmanned Aerial Vehicles
School Name
South Carolina Governor's School for Science and Mathematics
Grade Level
12th Grade
Presentation Topic
Computer Science
Presentation Type
Mentored
Abstract
The rise in computer-vision-related tasks in recent years calls for further development and research in object detection and its various applications across different fields of study. Object Detection typically refers to three primary tasks: detection, localization, and classification. This study focuses on training the YOLOv7 object detection model to detect objects along railways for railway maintenance using drones. For the study, the website “Roboflow” was used to gather and annotate 700 images of objects scattered along railways for the dataset. After implementing the YOLOv7 commands in the terminal, the algorithm splits the data into three folders for training, validation, and testing. After 28 hours of training, the results indicated that the algorithm was efficient in detecting, localizing, and classifying the objects. One issue with the results is with the validation objectness data, in which the inaccuracy rate increases as the number of epochs increases. In consequence, the algorithm struggled to box in the object in the validation dataset accurately. One cause could be the size of the dataset; with more training, the validation objectness inaccuracy rate should go down. Overall, the results indicate a successful training of the algorithm. In the future, new datasets can be trained by researchers or engineers for other applications relating to environmental conservation or security surveillance.
Recommended Citation
Zhu, Chloe, "Object Detection for Unmanned Aerial Vehicles" (2024). South Carolina Junior Academy of Science. 445.
https://scholarexchange.furman.edu/scjas/2024/all/445
Location
RITA 273
Start Date
3-23-2024 9:30 AM
Presentation Format
Oral Only
Group Project
No
Object Detection for Unmanned Aerial Vehicles
RITA 273
The rise in computer-vision-related tasks in recent years calls for further development and research in object detection and its various applications across different fields of study. Object Detection typically refers to three primary tasks: detection, localization, and classification. This study focuses on training the YOLOv7 object detection model to detect objects along railways for railway maintenance using drones. For the study, the website “Roboflow” was used to gather and annotate 700 images of objects scattered along railways for the dataset. After implementing the YOLOv7 commands in the terminal, the algorithm splits the data into three folders for training, validation, and testing. After 28 hours of training, the results indicated that the algorithm was efficient in detecting, localizing, and classifying the objects. One issue with the results is with the validation objectness data, in which the inaccuracy rate increases as the number of epochs increases. In consequence, the algorithm struggled to box in the object in the validation dataset accurately. One cause could be the size of the dataset; with more training, the validation objectness inaccuracy rate should go down. Overall, the results indicate a successful training of the algorithm. In the future, new datasets can be trained by researchers or engineers for other applications relating to environmental conservation or security surveillance.
