- New Deep Neural Network makes it easier for vehicles to see at night while ensuring other drivers won’t be blinded by the headlight.
- It dims individual high beam LEDs, creating glare-free zones as required by traffic patterns.
Although high-beam lamps incorporated into the front headlights increase night-time visibility range significantly, they can create dangerous glare to other drivers.
Today, all vehicles are installed with high beam headlights to give drivers bright, long-range illumination, but these light systems still require manual control. And since human drivers are prone to error, they mostly misuse (or underuse) these high beam lights.
Now, researchers at Nvidia Corporation have developed an artificial intelligence model that makes it easier for vehicles to see at night while ensuring other drivers won’t be blinded by the headlight. In other words, the AI uses perception to decrease glare for oncoming cars.
AutoHighBeamNet
In order to increase driving visibility in the dark without compromising the safety, researchers trained a deep neural network (DNN) on pictures captured by the vehicle’s front cameras. This camera-based DNN named AutoHighBeamNet learns from a wide range of scenarios for robust and fully automated headlight control.
The DNN responds to active automobiles (vehicles whose taillight or headlight is turned on) in the current camera frame. Cars parked on the side of the road, for instance, are inactive vehicles and they are ignored by the network.
System integration architecture
The DNN makes up the AutoDrivingBeam visual perception module that captures information from a monocular front camera. The post-processing submodule takes data from AutoHighBeamNet and processes each frame.
The auto OEM then customizes the outcomes of the AutoDrivingBeam module. It also takes other vehicle modules (such as environmental illumination conditions, speed of the ego car, etc.) into account.
Source: NVIDIA Developer | NVIDIA DRIVE Networks | YouTube
The high beam control works on two different modes:
- Auto High Beam: provides binary off/on control
- Adaptive Driving Beam: produces glare-free zones by precisely controlling individual high beam LEDs.
Auto High Beam
In the first mode, the high beam light system automatically turns on in poor driving conditions (in fog or poorly illuminated nighttime). When the system detects an active vehicle, it automatically switches to low beam or turns off. Once the active vehicle(s) get out of the camera frame, the high beams automatically turn back on.
Adaptive Driving Beam
In the second mode, the high beams are controlled in a way that prevents hazardous glare to active vehicles. The AI dims individual LEDs in the headlamp, creating glare-free zones as required by traffic patterns.
When active vehicles exit the camera frame, the dimmed zones are automatically turned backed to full luminance. Thus, this mode can always be kept in active mode for better safety.
Read: Self-Driving Vehicles Find It Hard To Detect Dark-Skinned Pedestrians
This technology was adopted in European countries a couple of years ago. It hasn’t yet been approved for the United States market, though it’s under active evaluation by the National Highway Traffic Safety Administration.
