Road Anomaly Dataset

Dataset Description

The RAD dataset comprises an extensive collection of high-quality images and videos, systematically annotated to reflect a variety of road anomalies. The dataset focuses on four primary categories, each representing a critical risk to human safety and property:

• Accidents: Includes footage of car crashes, multi-vehicle collisions, and other traffic-related incidents that pose immediate dangers.
• Car Fires: Captures instances of vehicles engulfed in flames, emphasizing the importance of quick detection to prevent further casualties and property damage.
• Fighting: Contains videos of physical altercations occurring on streets or roads, which could escalate into severe violence or endanger bystanders.
• Armed Robberies (Snatching at Gunpoint): Features videos of robberies and thefts involving firearms, where immediate response is essential for protecting lives and reducing the impact of criminal activities.

Each category in the dataset is annotated with detailed labels to ensure the accurate identification and classification of these anomalies. The annotations include temporal and spatial metadata, allowing for precise localization of the anomalies within the footage. This dataset is primarily aimed at training and testing machine learning models, particularly those using Convolutional Neural Networks (CNNs) and other deep learning architectures for high-accuracy anomaly detection.

Additional Categories and Data Enrichment

To further enhance the dataset’s utility, additional categories of road anomalies are being considered, including:

• Illegal Parking: Captures instances of vehicles parked in unauthorized or hazardous locations, potentially obstructing traffic or emergency services.
• Obstructions on Roadways: Identifies obstacles such as fallen trees, debris, or stalled vehicles, which can disrupt traffic flow and pose risks to motorists.
• Pedestrian Accidents: Includes footage of collisions involving pedestrians, which are often more severe due to the vulnerability of those involved.

Moreover, the RAD dataset has been enriched with contextually relevant metadata, such as environmental conditions (e.g., weather, lighting) and road types (e.g., urban streets, highways), further refining the data for more nuanced training applications.

Applications

The RAD dataset is ideal for developers and researchers working in public safety, smart cities, and autonomous vehicle technologies. Its primary applications include:

• Surveillance Systems: To improve the accuracy and responsiveness of AI-powered surveillance systems by detecting hazardous road activities.
• Traffic Management: To help authorities monitor and manage traffic flow more efficiently, identifying potential threats in real-time.
• Autonomous Vehicles: For the training of autonomous vehicle systems, equipping them to recognize and respond to road anomalies during navigation.