Self-driving cars have been a hot-button topic in recent years, particularly because they have the potential to improve passenger safety. Not only can they recognize other cars on the road, but they can also observe and react to pedestrians and even animals like birds and squirrels. But there’s a very popular entity they aren’t able to see so well: bicycles.
In fact, bikes are among one of the toughest challenges technological developers face in terms of detection. They’re smaller than cars and move quickly.
Nuno Vasconcelos, a visual computing expert at the University of California, San Diego, notes, “A car is basically a big block of stuff. A bicycle has much less mass and also there can be more variation in appearance — there are more shapes and colors and people hang stuff on them.”
The physical differences between bikes and cars is really throwing these vehicles’ detection systems for a loop. The algorithms on these systems train themselves based on the images they see. But many of the training sessions performed don’t feature bikes. As a result, the systems don’t consistently register them.
The Deep3DBox algorithm, for example, was able to recognize 89% of cars on the road when subjected to an industry-recognized benchmark test. But the system wasn’t nearly as proficient when it came to cyclists. In the test, it was able to spot only 74% of bicycles — and this score is the highest among such technologies.
With an increased emphasis on reducing carbon emissions and improving physical fitness, it’s no surprise that 3 million Americans age seven and up rode a bicycle at least six times in 2015. But if developers expect self-driving cars to become the standard, this technology will need to improve quickly.
Fortunately, companies are already starting to take action. High-definition maps, like the ones created by Israeli company Mobileye, can help these algorithms identifying bicycles by making them stand out against the landscape. Their Road Experience Management system could make huge improvements to the way these algorithms are trained. Google’s self-driving vehicle, Waymo, shows that improvements can certainly be made; that auto contains sensor technology with increased recognition for bicycles.
While some experts are skeptical that self-driving cars will ever completely replace human-operated ones, these improvements could definitely help ensure the safety of everyone on the road. That being said, the new technology needs to be carefully tested before going to market. Otherwise, drivers will be much better off relying on their own vision and skills to avoid an accident.