Dataset: Homography Image Rectification for Hazmat Sign Detection

MARS Hazmat Dataset

Paper: 
Improving CNN-based Planar Object Detection with Geometric Prior Knowledge; 
Jianxiong Cai, Jiawei Hou, Yiren Lu, Hongyu Chen, Laurent Kneip and Sören Schwertfeger; 
IEEE International Symposium on Safety, Security, Rescue Robotics (SSRR): IEEE Press, 2020. 

We provide a high resolution RGB-D hazmat dataset with 13 labels in this paper. It contains both 360 RGB images and 360 depth images with a resolution of 1280∗720. Ground truth label information of RGB images is also provided. 130 of the RGB and depth images contain only one of 13 types of hazmat label. Each of these image contains two backgrounds and five positions (top left, top right, center, bottom left, bottom right.) For the rest of tje 216 images, each image contains 13 types of hazmat labels. Nine different angles (±75◦, ±60◦, ±45◦, ±30◦, 0◦ ) with three distances (1.25m, 1.50m, 1.75m) are included in these images.

MARS-hazmat-dataset.zip (910 MiB)

Dataset format