Ahmed Shaami
There is no denying that facial recognition systems still struggle to perform efficiently when it comes to users with darker skin. Recently, a new report has emerged that shows that the systems that have been designed to help autonomous cars navigate roads will have similar problems. This means that pedestrians with darker skin tones could be in danger.
The problem:
Published in a newspaper, Researchers from the Georgia Institute of Technology investigated eight of the AI models that are used in state-of-the-art object detection systems. The systems allowed autonomous vehicles to identify road signs, pedestrians, as well as other objects. Images of pedestrians that have been used were divided into two categories on their score of the Fitzpatrick scale. This is the scale that is commonly used to classify the human color.
The study revealed that the models exhibited poor performance with the three darkest shades on the scale. While examining the group containing images of pedestrians with darker skin tones, the models’ accuracy lowered to 5 percent.
The solution:
After the investigation, it was suggested that autonomous cars are not safe for people with darker skin tone as compared for their lighter-skinned pedestrians. They stated on the basis of their study that what should be done to avoid a future of biased self-driving cars. The process can be made better by including more images of dark-skinned pedestrians in the data sets of the systems.
There is no denying that facial recognition systems still struggle to perform efficiently when it comes to users with darker skin. Recently, a new report has emerged that shows that the systems that have been designed to help autonomous cars navigate roads will have similar problems. This means that pedestrians with darker skin tones could be in danger.
