DAVIS Dayton Annotated Vehicle Image Set

An Annotated Vehicle Image Set from Data Collected by Roadside Traffic Cameras 

Rationale for Creating This Dataset

Consummating a demand from the National Perspective, a model that has been designed for real life application of accurate vehicle recognition and classification in respect of US roads, based upon a wide range of traffic scenarios, vehicle types, sizes, orientations and environmental conditions specific to the United States whose operation is supposed to be effective upon deployment. This claims for investiture of a specific dataset for better functionality of machine learning algorithm training to be dependable on US roads.

Details Specification of the dataset

A dataset comprising of 914 training images with their related annotations and 249 validation images, all carefully labelled with matching labels is the constitution of The Dayton Annotated Vehicle Image Set (DAVIS) which is carefully crafted to provide warranty of the reliability and efficiency of Vehicle detection on US roads. Capturing images through real-time security camera and footage from various organizations like Gebhardt Insurance Company, SharX security, and the Village of Tilton's traffic camera system and feeding with diverse inputs is the foundation upon which this Dataset is ensconced. This extensive dataset is a repertoire of a variety of viewpoints on traffic scenarios, exhibiting a broad spectrum of vehicle sizes, orientations and types on US roadways. Precision in vehicle demarcation within each image has been derived through application of Open-source software to assist annotation and label deployment. This well-positioned Dataset crafted upon sophisticated annotation tools and integration of additional internet-sourced images with real-world video footage, enabled vehicles in various scenarios and environmental conditions on US roads serves aa a solid training base for the YOLOv8 network.

Figure: The left image represents the original frame captured by the roadside traffic camera. The center image shows the corresponding annotated version highlighting detected vehicles. The right image displays the YOLOv8 annotation file containing the class code along with the bounding box coordinates (top left, bottom right).

Data Download

To download the dataset, click here: https://docs.google.com/forms/d/e/1FAIpQLScobrzZmLqs8p1HxdAbqWx9L1TcJbAFNEj6d88Osjs3Lm--0g/viewform

The training dataset includes five object classes. Class labels used for training are defined as follows:

  0: 'cars'

  1: 'motorcycles'

  2: 'trucks'

  3: 'buses'

  4: 'workvans'

Acknowledgement

The source images are collected from the live stream data provided by Gebhardt Insurance Company, SharX Security, and Village of Tilton's Traffic Camera System. 

Contact

If you used our DAMIE data set we would love to hear from you! Please consider sending us a message about your results and any feedback you have.
For additional questions about this data set or inquiries about commercial use, please contact  mallika1@udayton.edu