Study of computer vision methods for identifying obstacles on forest roads

Methods of technical and computer vision are developing and being increasingly used in various civil industries. Computer vision tools may be used to improve passability and traffic safety of forest machinery by promptly recognizing isolated obstacles in the form of roots, stumps, hummocks, potholes, etc. The study tested methodology of an experiment to study capabilities of computer vision tools for recognizing isolated obstacles on forest roads. The experiments in were carried out for various versions of the YOLO artificial neural network (YOLOv8n.pt, YOLOv8s.pt, YOLOv8m.pt, YOLOv8l.pt), retrained on a large dataset of Road Damage Detection 2022. It was found that the experimental setup, including software and hardware, as well as the selected hyperparameters of the model training process, make it possible to obtain stable experimental data on the recognition and classification of road defects, including forest ones. The results of scoring the YOLO models during retraining and validation showed that the YOLOv8m.pt artificial neural network model should be recommended as a promising version for developing a technical solution for recognizing single obstacles on forest roads; however, the issue of regularizing the model weights should be additionally considered. Testing and expert evaluation of the results confirmed the preliminary conclusions about the promise of the YOLOv8m.pt version as basis for the technical solution. The expediency of using the numerical optimization method Adam with a minimization step of 0.00001 in further studies related to experiments with the models of the artificial neural network YOLOv9, YOLOv9v10 is noted for the purpose of compiling a more complete and systematic scientific understanding of the applicability of computer vision models for identifying isolated obstacles on forest roads.

1. Han S., Jiang X., Wu Z. An Improved YOLOv5 Algorithm for Wood Defect Detection Based on Attention. IEEE Access, 2023, vol. 11, pp. 71800–71810. URL: https://api.semanticscholar.org/CorpusID:259721127 (accessed August 10, 2024).

2. Hutter F., Kotthoff L., Vanschoren J. (editors). Automated Machine Learning: Methods, Systems, Challenges. Springer, 2019. DOI: 10.1007/978-3-030-05318-5.

3. Latour B. Science in Action: How to Follow Scientists and Engineers through Society. Cambridge, MA: Harvard University Press, 1987.

4. Mohan Prakash B., Sriharipriya K.C. Enhanced pothole detection system using YOLOX algorithm. Auton. Intell. Syst, 2022, 2, 22. URL: https://doi.org/10.1007/s43684-022-00037-z

5. Pothole Dataset: Project on Realtime Potholes Detection. URL: https://public.roboflow.com/object-detection/pothole (accessed August 10, 2024).

6. RDD2022 Dataset: The Multi-National Road Damage Dataset 2022. URL: https://datasetninja.com/road-damage-detector (accessed August 08, 2024).

7. Shevtekar S. Enhanced Pothole Detection Using YOLOv8 Nano. International Scientific Journal of Engineering and Management, 2024, 03, pp. 1–9. 10.55041/ISJEM01632.2024

8. Wang M., Li M., Cui W., Xiang X., Duo H. TSW-YOLO-v8n: Optimization of detection algorithms for surface defects on sawn timber. BioResources, 2023, vol. 18, no. 4, pp. 8444–8457.

9. Wang R., Chen Y., Liang F., Wang B., Mou X., Zhang G. BPN-YOLO: A Novel Method for Wood Defect Detection Based on YOLOv7. Forests. 2024. URL: https://api.semanticscholar.org/CorpusID:270753559 (access August 08, 2024). Wang R.,

10. Liang F., Wang B., Mou X. ODCA-YOLO: An Omni-Dynamic Convolution Coordinate Attention-Based YOLO for Wood Defect Detection. Forests, 2023, vol. 14, no. 9, 1885 p. DOI: https://doi.org/10.3390/f14091885.

11. YOLOv8 Docs: Ultralytics YOLO Docs. URL: https://docs.ultralytics.com (accessed August 08, 2024).