New AI Breakthrough Lets You See Around Corners in Real Time
Imagine your phone or car seeing around blind spots instantly — this tech could do that.
Non-line-of-sight imaging can reconstruct scenes hidden around a corner from indirect light recorded by a single-photon avalanche diode. Researchers rebuilt GPU execution of two wave-based algorithms—f-k migration and phasor-fields—with memory-saving techniques like offline ring kernels, fused kernels, warp-level photon binning, CUDA graph replay, and selective FP16 storage. Their implementations run up to 42x faster than a reference streaming pipeline and up to 14x faster than the fastest published GPU baseline, while using as little as 2.5% of the memory—enabling much larger and finer reconstructions on the same hardware, and they outline three denoising strategies for next-generation NLOS video processing.
- New AI method lets cameras see hidden objects by analyzing light bouncing off walls — like X-ray vision for everyday tech.
- It’s 42 times faster than before and runs on ordinary hardware, making it possible to use in phones or cars.
- Potential uses include safer self-driving cars, better emergency rescue tools, and even helping you find lost items at home.
Why It Matters
This tech could make your future phone or car see around corners, improving safety and convenience in daily life.