May 21-23, 2024
Huntington Place, Detroit
Thermal Imaging (LWIR) is proving to be an important sensing modality in visually degraded environments from bright-light scenarios to no-light scenes. Previously, the challenge for deployment has been the cost of thermal vs. visible light cameras and the lack of camera resolution. As a result, visible light cameras are being pressed to their limits to perform in the most challenging use cases and fail in complete darkness. Thermal imaging advancements are making this modality a viable solution.
Join us as Owl will address in detail upcoming regulatory mandates, specifically from the National Highway Traffic Safety Administration and EURO NCAP for nighttime pedestrian safety. Between 2010 and 2021, pedestrian fatalities in the US increased by 77%, the majority occurring at night according to the Governors Highway Safety Association.
The architecture of thermal cameras has changed in recent years. Owl will highlight the architectural changes that are enabling cost-effective, high-resolution, low-power next-generation cameras that will be embraced by automotive OEMs for ADAS and autonomy deployments.
Current systems have been relying on a combination of visible light cameras and radar to identify VRUs.
To help participants better understand the capabilities of HD thermal, this talk will:
Show real-world comparison of visible light cameras to LWIR thermal in degraded visual environments (DVEs) such as:
- Nighttime driving
- Bright oncoming sunlight
- Heavy fog
- Rain
- Snow
- Chaotic urban environments
Thermal imaging has been plagued by high-cost architectures and challenges of shutterless calibration. The Owl talk will show a path to low-cost thermal cameras and highlight progress towards addressing the technical barriers to entry in the automotive market.
Computer vision in thermal imaging has been slow to be adopted as training datasets have been unavailable. Owl will demonstrate a new dataset designed for improved classification, segmentation and fusion to accelerate automotive OEM time to market. Benchmarks comparing various neural networks will be presented.
