Who is interested to use this data set for his research, please refer to this reference:
A Pashaei, M Ghatee, H Sajedi, Convolution neural network joint with mixture of extreme learning machines for feature extraction and classification of accident images, Journal of Real-Time Image Processing 17 (2020) 1051–1066.
This dataset has been collected to evaluate the performance of the different classifiers on the accident images. The images have been collected by UAV and some resources of Google images from different locations in various time. The size of these images were transformed to 28*28 pixels. They are labeled by human experts. This dataset includes three folders:
- Accident –Detection
The first folder: Accident –Detection
The aim of this dataset is to detect the occurrence of accidents, by image processing. It includes two subfolders with labels “without-accident” and “with-accident”.
- Folder 1 includes 2500 images with label “without-accident”.
- Folder 2 includes 2398 images with label “with-accident”
The second folder: Vehicles-in-Accidents
The aim of this dataset is to classify the vehicles involved in an accident, by image processing. It includes three subfolders with labels “light vehicle”, “heavy vehicle” and “motorcycle”.
- Folder 1 includes 892 images with label “light vehicle”.
- Folder 2 includes 876 images with label “heavy vehicle”.
- Folder 3 includes 868 images with label “motorcycle”.
The third folder: Accident-Severity
The aim of this dataset is to classify the severity of accidents, by image processing. It includes three subfolders with labels ” low dangerous”, ” medium dangerous” and ” high dangerous”.
- Folder 1 includes 118 images with label “low dangerous”.
- Folder 2 includes 1603 images with label “medium dangerous”.
- Folder 3 includes 1225 images with label “high dangerous”.
The graphical shape of this dataset is given in the following figure:
To download this dataset, click here. To download the original figures that have been used to produce this dataset can be shared by emailing to authors.