ADMET

Complete News World in United States

Self-driving cars have trouble recognizing darker skin tones and people with disabilities

Self-driving cars have trouble recognizing darker skin tones and people with disabilities

The UK Legal Commission has begun developing a legal framework for the introduction of autonomous vehicles on UK roads. A 2019 report showed that vehicles that use automatic-driving technologies may have problems recognizing dark-skinned faces in the dark.

It is estimated that around 33 million autonomous vehicles will be on the roads by 2040. Although this number is impressive, the level of development and refinement of algorithms responsible for safe driving remains in question. Research by scientists clearly shows some issues with the correct recognition of skin tone.

Georgia Tech researchers have investigated eight AI models used in the latest object detection systems. These systems allow self-driving vehicles to recognize road signs, pedestrians, and other objects as they move on the roads.

The Fitzpatrick scale was used to determine the accuracy of algorithms – a numerical scheme for classifying human skin color. System accuracy decreased an average of 5%. At the time the darker shades were checked. The models performed “uniformly poorly” when confronted with pedestrians in three shades of darkness on the scale.

The situation was similar in the case of, for example, people with disabilities in wheelchairs. Artificial intelligence systems have had a problem with the precise allocation of such a person to a particular class of recognizable objects on the road.

However, scientists agree that these problems can be solved. Autonomous systems used in cars should be uploaded to broader sets of images, matched to a specific skin tone, or vehicles designated for the movement of a person with a disability.

READ  Bonanza in a state-owned company. The father will supervise the son? O2

So engineers responsible for developing these solutions should focus more on training systems more precisely for a particular group.