A Performance Comparison of Object Detection Algorithms on Traffic Scenes in Indian Roads
Saved in:
| Title: | A Performance Comparison of Object Detection Algorithms on Traffic Scenes in Indian Roads |
|---|---|
| Authors: | Bhakti Paranjape, Apurva Naik, S. Perumal Sankar |
| Source: | Engineering, Technology & Applied Science Research. 15:25492-25498 |
| Publisher Information: | Engineering, Technology & Applied Science Research, 2025. |
| Publication Year: | 2025 |
| Description: | Machine learning-based object detection allows machines to decipher visual information and recognize objects in digital images or videos using localization and classification techniques. This study focuses on applying object detection techniques to traffic images from Indian roads, which are unstructured and have complicated traffic patterns. Taking into account the high number of traffic accidents in India, it is imperative to develop intelligent systems for traffic analysis and management. This study uses four cutting-edge object detection algorithms, SSD, YOLO, Faster R-CNN, and CenterNet, previously trained on the popular COCO and PASCAL VOC datasets. These algorithms are tested on the DATS dataset, created to represent road conditions in India, to examine the capacity of these models to manage the complexities of Indian traffic situations. In terms of mAP, the results show that CenterNet had the lowest score (69.5%) and YOLOv3 the highest (81.5%). |
| Document Type: | Article |
| ISSN: | 1792-8036 2241-4487 |
| DOI: | 10.48084/etasr.11105 |
| Rights: | CC BY |
| Accession Number: | edsair.doi...........1ef6e10a2713ab19c9c2602912ab194f |
| Database: | OpenAIRE |
Nájsť tento článok vo Web of Science | Abstract: | Machine learning-based object detection allows machines to decipher visual information and recognize objects in digital images or videos using localization and classification techniques. This study focuses on applying object detection techniques to traffic images from Indian roads, which are unstructured and have complicated traffic patterns. Taking into account the high number of traffic accidents in India, it is imperative to develop intelligent systems for traffic analysis and management. This study uses four cutting-edge object detection algorithms, SSD, YOLO, Faster R-CNN, and CenterNet, previously trained on the popular COCO and PASCAL VOC datasets. These algorithms are tested on the DATS dataset, created to represent road conditions in India, to examine the capacity of these models to manage the complexities of Indian traffic situations. In terms of mAP, the results show that CenterNet had the lowest score (69.5%) and YOLOv3 the highest (81.5%). |
|---|---|
| ISSN: | 17928036 22414487 |
| DOI: | 10.48084/etasr.11105 |