| Title |
Influence of aerial image resolution on vehicle detection accuracy / |
| Authors |
Gliaubičiūtė, Donata ; Janavičius, Rokas ; Gadeikytė, Aušra ; Paulauskas, Lukas |
| Full Text |
|
| Is Part of |
CEUR workshop proceedings: IVUS 2023: Information society and university studies 2023: proceedings of the 28th international conference on information society and university studies (IVUS 2023) Kaunas, Lithuania, May 12, 2023 / edited by: A. Lopata, T. Krilavičius, I. Veitaitė, A. García-Holgado.. Aachen : CEUR-WS. 2023, vol. 3575, p. 152-161.. ISSN 1613-0073 |
| Keywords [eng] |
vehicle detection ; aerial images ; convolutional neural networks ; pixel ratio ; YOLOv5 ; YOLOv7 ; YOLOv8 |
| Abstract [eng] |
Nowadays, the engineering application of vehicle detection from aerial images is a challenging task due to the particularity of perspective, the small size of the objects, and the complex background. This research aim is to investigate low-resolution aerial images of vehicles that can be utilized for vehicle detection using machine learning models. The research work was conducted using one-stage deep learning-based object detection algorithms YOLOv5, YOLOv7, and YOLOv8 on the two datasets (COWC and VEDAI) that addressed the task of small vehicle detection. For the training of the models, available pre-trained weights were used as a starting point, and then each model was trained by utilizing transfer learning. The obtained results of the study demonstrated that by reducing the image pixel ratio every 5 cm per pixel from 12.5x12.5 to 27.5x27.5 cm per pixel, the accuracy of the object detection models decreases by an average of 3.51%. When the pixel ratio varies from 30x30 to 32.5x32.5 cm per pixel, the accuracy of the models drops by an average of 2.33% on the COWC dataset and 42.4% on the VEDAI dataset. |
| Published |
Aachen : CEUR-WS |
| Type |
Conference paper |
| Language |
English |
| Publication date |
2023 |
| CC license |
|
