Comparative Study of Different Classification Techniques for Pedestrian Detection Application
| Autor: | Tuqa Hani Abd-Alamir, Mohammed Sadoon Hathal |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2024 |
| Předmět: | |
| Zdroj: | Journal of Engineering, Vol 30, Iss 8 (2024) |
| Druh dokumentu: | article |
| ISSN: | 1726-4073 2520-3339 |
| DOI: | 10.31026/j.eng.2024.08.10 |
| Popis: | Pedestrian detection is well known as one of the most important applications in computer vision. However, reliable pedestrian detection is difficult due to a variety of factors, including changing size of pedestrian characteristics and crowded backgrounds. This study aims to evaluate and compare the pedestrian detection performance of three different types of classifiers: Random-Forest (RF), Convolution-Neural-Network (CNN), and Support-Vector-Machine (SVM). The presented methodology involves using You_Only_Look_Once (YOLOv8) architecture for object segmentation and the Histogram of Oriented Gradients (HOG) for feature extraction. Then, RF, CNN and SVM classifiers are trained and tested using the extracted HOG features. Using the EPFL pedestrian dataset, the experiment showed that the CNN model returned the highest results which had a speed of 0.42s and an accuracy percentage of 93.34%. Compared to SVM and RF, CNN provides a high detection speed and accuracy. RF has the slowest detection speed, while SVM has the lowest detection accuracy. This study gives useful information regarding the efficacy of these classifiers in detecting pedestrians under various weather circumstances, and the findings show that CNNs can achieve high accuracy while maintaining remarkable detection efficiency. |
| Databáze: | Directory of Open Access Journals |
| Externí odkaz: |
|