VGU RESEARCH REPOSITORY
Please use this identifier to cite or link to this item:
https://epub.vgu.edu.vn/handle/dlibvgu/1864| Title: | Labeling detection dataset with detection and tracking models | Authors: | Nguyen Quoc Huy | Keywords: | Object detection;Tracking model;Traffic sign detection | Issue Date: | 2024 | Abstract: | This study will analyze different patterns of traffic signs inside an image to classify and label each type of traffic sign into a specific category or class. The purpose of this research is to enhance the performance of the YOLOv8m object detection model by applying an Active Learning method called Diversity Sampling to extract a new dataset from GoPro videos. By investigating error percentages through a Tracking Algorithm, we aim to reduce errors in the detection and classification stages. Our findings indicate that while training errors have been significantly diminished, the model still encounters substantial errors when classifying new real-time data. This discrepancy is largely attributed to environmental factors and similar obstacles that challenge the model's accuracy. Our experiments also reveal a strong correlation between model performance and the accuracy and volume of the training dataset-demonstrating that more accurate and extensive datasets yield more reliable models. However, a limitation of this study is that the correct class ID for traffic signs is determined by the frequency of appearance, with the assumption that the most common class ID is the appropriate label. |
URI(1): | https://epub.vgu.edu.vn/handle/dlibvgu/1864 | Rights: | Attribution-NonCommercial 4.0 International |
| Appears in Collections: | Computer Science (CS) |
Files in This Item:
| File | Description | Size | Format | Existing users please Login |
|---|---|---|---|---|
| Labeling detection dataset with detection and tracking models.pdf | 11.27 MB | Adobe PDF |
Page view(s)
55
checked on Oct 30, 2025
Download(s)
16
checked on Oct 30, 2025
Google ScholarTM
Check
This item is licensed under a Creative Commons License