SMART VEHICLE DETECTION SYSTEM USING DEEP LEARNING ARCHITECTURES

Description

The Smart Vehicle Detection System Dataset is a multi-class image dataset developed for research in computer vision, deep learning, and intelligent transportation systems. It is designed primarily for vehicle type classification and can be used for benchmarking machine learning models, transfer learning, and related computer vision applications. The dataset contains over 5,000 RGB images spanning more than 30 vehicle categories, including cars, motorcycles, buses, trucks, vans, minibuses, pickup trucks, three-wheelers, bicycles, emergency vehicles, and other road vehicles. Each image is assigned a single class label representing the primary vehicle type. ## Data Collection Images were collected from two sources: photographs captured in real-world traffic environments using multiple smartphone cameras and publicly available images from academic and open-source repositories. The dataset includes diverse lighting conditions, viewing angles, backgrounds, and vehicle appearances to improve model robustness. ## Image Processing Images were manually reviewed to remove duplicates, blurry samples, and low-quality images. Each image was cropped, resized to 224 × 224 pixels, converted to RGB, and saved in JPEG (.jpg) format. Labels are organized into class-specific folders and are also provided in a metadata.csv file. ## Dataset Structure The dataset follows a hierarchical folder structure compatible with common deep learning frameworks such as PyTorch, TensorFlow, and Keras, enabling direct use for image classification tasks. ## Applications This dataset supports research in: * Vehicle type classification * Deep learning and computer vision * Intelligent transportation systems * Traffic monitoring * Smart city applications * Transfer learning and model benchmarking It can also be extended for object detection and related vision tasks through additional annotations. ## Limitations The dataset primarily represents vehicle types common to the collection region and may not fully generalize to all geographic locations. It contains only still RGB images and does not include bounding-box annotations, video sequences, depth information, or nighttime and adverse-weather images.

Institutions

• Southeast University

  Dhaka Division, Dhaka

Categories

Licence