We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Research on Pedestrian Detection AlgorithmBased on Deep Learning.
- Authors
Ying Wang; Ying Tian
- Abstract
Pedestrian detection model based on deep learning has been widely used in various fields. However, its capabilities are limited when operating in complex environments due to high false alarm rates and low detection accuracy. To address these limitations, this paper presents an improved model called YOLOv5s-FCC. This enhanced model retains the distinctive features of the original model while improving several aspects such as the loss function, feature extraction, feature output, and anchor frames. The model's performance is evaluated on the CrowdHuman and WiderPerson datasets, which consist of high pedestrian density and significant obstacles, making detection challenging. To overcome these challenges, the datasets are first re-clustered using the K-Means++ clustering technique to obtain optimal anchor frames. Additionally, the Focal-EIOU loss function is employed to accelerate convergence speed and improve regression results. A coordinate convolution layer is fused before the output of different-scale features to provide more informative content. Lastly, a new pyramid pool layer of CSPCSPPF space is used to extract feature information, which reduces the repetition of gradient information and realizes more accurate detection. Comparative experiments using testing sets from both datasets demonstrate an improvement in Mean Average Precision of 2.2% and 1.8%, respectively, along with a reduction in the miss rate of 4.6% and 3.7%.
- Subjects
DEEP learning; PEDESTRIANS; FEATURE extraction
- Publication
IAENG International Journal of Computer Science, 2023, Vol 50, Issue 4, p1446
- ISSN
1819-656X
- Publication type
Article