Development of a vision-based bus-passenger data counting and density mapping for GPS location clustering
Bus intelligent systems are an important aspect of improving the existing systems to understand and gather information. Informed decisions then could be made using the system. The information system is comprised of a vision-based bus passenger counter, a bus density mapper, and a GPS tracker to anal...
Saved in:
| Main Author: | |
|---|---|
| Format: | text |
| Language: | English |
| Published: |
Animo Repository
2021
|
| Subjects: | |
| Online Access: | https://animorepository.dlsu.edu.ph/etdm_ece/8 https://animorepository.dlsu.edu.ph/context/etdm_ece/article/1007/viewcontent/velasco_Redacted.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Bus intelligent systems are an important aspect of improving the existing systems to understand and gather information. Informed decisions then could be made using the system. The information system is comprised of a vision-based bus passenger counter, a bus density mapper, and a GPS tracker to analyze its locations and passenger count data. The bus counter used Scaled YOLOv4 and FastMOT tracking in counting in 3 lighting conditions from PCDS dataset. The density mapper used Bayesian Crowd Counting model with a Mish activation trained with mall dataset to display and determine density. Lastly, the GPS data is processed with T-DBSCAN data clustering to display relevant stops and trajectories. The processes are simulated using open-sourced data to log to a database for storage and analysis. Passenger counter performs at 88.00%, 83.78%, 90.32% at normal, noisy, and night conditions. Bayesian VGG-Mish video heatmap is visually better compared to using ReLU and Swish. Lastly, the GPS clustering, although not contextualized, is able to output trajectory stops with its arrival time and departure at the location. |
|---|