Research on the impact of Non_Maximal Suppression threshold value on YOLO’s ability to recognize objects in low-quality images

Authors

DOI:

https://doi.org/10.18372/2073-4751.74.17884

Keywords:

YOLO, YOLOv7, UAV, NMS, objects detection

Abstract

Based on the analysis of existing approaches to object detection in images, the YOLO model was selected for investigating the dependency of performance on the Non-Maximal Suppression (NMS) threshold value. This article addresses the current challenge of utilizing unmanned aerial vehicles (UAVs), particularly drones, in various fields where they are employed for collecting visual data. The primary objective of this research is to study and evaluate the optimal NMS threshold value for the YOLOv7 model when processing images acquired from UAVs under conditions characterized by low resolution, noise, and other artifacts. The study demonstrates that the YOLOv7 model can be effective in recognizing objects in images obtained from drones, even in the presence of low resolution and noise. However, it was observed that altering the NMS parameter affects the accuracy and frequency of false object detections. Decreasing the parameter value can enhance object recognition but concurrently increases the likelihood of false detections. The obtained results indicate the need for further research in this area, including improving the quality of source images, developing individualized approaches, and methods for working with visual data obtained from UAVs with low resolution.

References

Jonathan Hui. Object detection: speed and accuracy comparison (Faster R-CNN, R-FCN, SSD, FPN, RetinaNet and YOLOv3). URL: https://jonathan-hui.medium.com/object-detection-speed-and-accuracy-comparison-faster-r-cnn-r-fcn-ssd-and-yolo-5425656ae359.

Chen C., Zheng Z., Xu T., Guo S., Feng S., Yao W., Lan Y. YOLO-Based UAV Technology: A Review of the Research and Its Applications. Drones. 2023. V. 7. 190.

Buchholz K., Statista. Commercial Drones Are Taking Off. URL: https://www.statista.com/chart/17201/commecial-drones-projected-growth/.

Yang H., Liu Y., Wang S., Qu H., Li N., Wu J., Yan Y., Zhang H., Wang J., Qiu J. Improved Apple Fruit Target Recognition Method Based on YOLOv7 Model. Agriculture. 2023. V. 13. 1278.

Weber E., Vedaldi A., Bischof H., Brox T., Frahm J.M. Detecting Natural Disasters, Damage, and Incidents in the Wild. Computer Vision – ECCV 2020. ECCV 2020. Lecture Notes in Computer Science. 2020. V. 12364. P. 331–350.

Du D., Qi Y., Yu H.g, Yang Y., Duan K., Li G., Zhang W.g, Huang Q., Tian Q. The Unmanned Aerial Vehicle Benchmark: Object Detection and Tracking. European Conference on Computer Vision (ECCV). 2018. 17 p.

Published

2023-06-30

Issue

Section

Статті