The team processed 8,000 images with the software and found that the systems were 20 percent better at detecting adult-sized pedestrians as opposed to children. They were also found to be 7.5 percent better at detecting light-skinned pedestrians than those with darker skin. “Car manufacturers don’t release the details of the software they use for pedestrian detection, but as they are usually built upon the same open source systems we used in our research, we can be quite sure that they are running into the same issues of bias,” said Dr. Jie Zhang, one of the authors of the study.
Are driverless cars less effective at spotting kids and people of color?
Answer: It appears so.
It’s no secret that many algorithms can be biased based on the data that they’re trained on, and it seems self-driving cars are no exception. A study by King's College London found that pedestrian detection systems used in autonomous driving research were less accurate when it came to identifying kids and people with darker skin.
The team processed 8,000 images with the software and found that the systems were 20 percent better at detecting adult-sized pedestrians as opposed to children. They were also found to be 7.5 percent better at detecting light-skinned pedestrians than those with darker skin. “Car manufacturers don’t release the details of the software they use for pedestrian detection, but as they are usually built upon the same open source systems we used in our research, we can be quite sure that they are running into the same issues of bias,” said Dr. Jie Zhang, one of the authors of the study.
A spokesperson from Waymo pushed back against the study’s findings, noting that self-driving cars use more than just cameras to determine their surroundings. “We tap into our full sensor suite — including our lidars and radars, not just cameras — to help us actively sense details in our surroundings in a way that would be difficult to do with cameras alone,” said Sandy Karp.
The team processed 8,000 images with the software and found that the systems were 20 percent better at detecting adult-sized pedestrians as opposed to children. They were also found to be 7.5 percent better at detecting light-skinned pedestrians than those with darker skin. “Car manufacturers don’t release the details of the software they use for pedestrian detection, but as they are usually built upon the same open source systems we used in our research, we can be quite sure that they are running into the same issues of bias,” said Dr. Jie Zhang, one of the authors of the study.
