'Smart headlights' could improve nighttime driving visibility
Nearly half of all car accidents occur at night despite fewer drivers on the road, according to the National Highway Safety Transportation Administration. Rain and snow only make matters worse, further reducing visibility. However, a team of Carnegie Mellon researchers are developing a way around this problem: smart headlights.
A vehicle’s headlights illuminate everything in front of the car. This means that when it’s raining or snowing, the headlights illuminate raindrops or snowflakes, which distort the light and prevent drivers from seeing objects as easily.
The researchers, led by associate professor Srinivasa Narasimhan of the Robotics Institute, solved this problem by streaming light between raindrops so it’s not reflected back toward the driver. Instead of using a bulb as a headlight, the researchers used a projector, which controls the pixels.
“If the particles are not hit by the projector light, the driver will not see them,” Narasimhan said.
The smart headlights have a camera that captures images illuminated by the light source, showing where the raindrops are. This information is transmitted to a computer, which figures out where the raindrops will be in the next few milliseconds by looking at their velocity and direction. The findings are then fed back to the headlights, which switch off light rays at those predicted locations.
Narasimhan’s team tested the headlights by writing a computer system that simulated all kinds of rain and snow conditions. After doing a feasibility study, the researchers built a real system in a laboratory. They used valves to generate real rain conditions, ranging from a drizzle to a downpour.
The project started about two years ago, and Narasimhan said that it may take another two or three years to test the system and make sure that it works for cars traveling at high speeds. Companies in the automotive industry have expressed interest in commercializing this product, Narasimhan said.
Currently, the smart headlights work for a car moving at around 20 miles per hour. In order to test the system on a faster moving car, there needs to be a significant improvement in the efficiency of data transfer between all of the headlight system’s components.
“For that, we are trying to build a completely embedded system,” Narasimhan said.
The bottleneck of the current system comes from the time it takes for an image to be transferred among the camera, computer, and projector — a transfer delay of eight milliseconds. Narasimhan explained that the team wants to put the camera and the processing on the same computing board with the projector directly connected.
“A couple of years ago, when we started this project, we thought it was going to be very hard,” Narasimhan said. Before the development of the current system, the team decided to start with an easier project, building a multi-dimensional display that simulated water drops falling through the air.
The prototype system had up to four layers, with each layer consisting of a row of fifty drops. The researchers knew exactly where the water drops were because they controlled them, which created a much simpler scenario than reality. Once the computational requirements became clear with the prototype system, the team was ready to use the smart headlight system to tackle the challenges associated with real, non-simulated rain drops.