AUTOMATED CLASSIFICATION AND COUNTING OF VEHICLES USING DEEP LEARNING APPROACH
The rapid increase in automobile traffic in and around cities presents significant challenges for transportation management, infrastructure planning, and overall system viability. Conventional vehicle classification and counting techniques, which are mostly manual or sensor-based, are often constrai...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
Little Lion Scientific
2025
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/51068/2/AUTOMATED.pdf http://ir.unimas.my/id/eprint/51068/ https://www.jatit.org/volumes/Vol103No23/10Vol103No23.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The rapid increase in automobile traffic in and around cities presents significant challenges for transportation management, infrastructure planning, and overall system viability. Conventional vehicle classification and counting techniques, which are mostly manual or sensor-based, are often constrained by inefficiency, labor intensity, and limited scalability in real-time applications. This paper proposes a deep learning approach for real-time classification and enumeration of vehicles using one-dimensional signals from piezoelectric sensors. The 1D-CNN was trained, using convolutional layers with small kernel sizes to stack, so that the temporal dependencies of sensor signals could be well learned, pooling and fully connected layers used to extract features with high strength. The model was trained using a hybrid dataset with field-collected sensor data and synthetically generated signals generated with traffic simulation tools to cover a classes of vehicles and road conditions, which guarantees the ability to scale and generalization. The proposed model is shown to perform well as it has a classification accuracy of 0.99, and mean Average Precision (mAP) of 0.98 on five different vehicle classes. Moreover, its performance was checked with the unseen data, which proved high generalization ability. The solution has potential to be deployed on edge computing devices because it is computationally efficient to support a realistic application of traffic monitoring. |
|---|