Context

This dataset was collected to train an SVM neural network to be shipped to help visually impaired people. Paper available in https://doi.org/10.1016/j.micpro.2020.103086

Content

This dataset has been divided into 4 classes, each class corresponding to a similar perspective set. The first class are crosswalks viewed from the front, the second and third correspond to the half-lane viewed from left or right, the fourth class is from non-crosswalk images, ie asphalt images, from passing cars and sidewalks. This set of images was extracted from several videos recorded with a resolution of 1280x720 at 30 FPS, in the city of Fortaleza-CE, Brazil during the period with sunlight. The images have the same resolution as the videos, 1280x720.

About file 10_FEATURES_M17_CM6b_TH199.csv

Example of the data set used on the embedded platform with the following configuration:

• Threshold images with T = 199
• Images resized by decimation with factor M = 17
• 10 GLCM features selected by complexity analysis.
• Co-occurrence matrix considering only 6 bits of grayscale images

Acknowledgements

This dataset was created by the authors themselves with images obtained by them.

Inspiration

What preprocessing techniques can be and what machine learning algorithm can be used to achieve good accuracy and sensitivity results with the lowest possible memory and computation time?