How Non-Line-of-Sight Imaging Sees Hidden Objects

This article provides a quick overview of how Non-Line-of-Sight (NLOS) imaging allows cameras to see around corners by combining ultra-fast laser physics and advanced computing. You will learn how femtosecond laser pulses act as a light source, how time-of-flight sensors capture scattered photons, and how transient rendering artificial intelligence processes this data to reconstruct hidden three-dimensional environments.

To see an object that is blocked from direct view, NLOS imaging systems must use a nearby visible surface, such as a wall or a floor, as a relay mirror. Because standard walls are rough and scatter light in all directions, normal cameras only see a blurry glow. NLOS overcomes this by firing femtosecond laser pulses at the visible wall. A femtosecond is one quadrillionth of a second. Firing a laser at this speed creates an ultra-short burst of light that hits the wall, scatters around the corner, hits the hidden object, and bounces back to the wall before returning to the camera sensor.

Capturing this returning light requires extreme precision, which is achieved through time-of-flight (ToF) technology. The NLOS system uses specialized detectors, often Single-Photon Avalanche Diodes (SPADs), alongside the laser. These sensors do not take a normal photograph; instead, they act as high-speed stopwatches. They measure the exact time it takes for individual photons from the femtosecond laser to complete their multi-bounce journey. Because the speed of light is constant, the system can calculate the exact distance the light traveled down to the millimeter.

Once the time-of-flight data is collected, it exists only as a chaotic cloud of photon arrival times. To turn this raw data into a recognizable picture, the system uses transient rendering and artificial intelligence. Transient rendering is a method that simulates how light travels and scatters through a scene over ultra-short time intervals, rather than assuming light travels instantly.

Artificial intelligence algorithms, specifically deep neural networks, are trained on these transient rendering models to solve the "inverse problem." The AI looks at the timing and pattern of the scattered photons that returned to the sensor and works backward. It filters out background noise, calculates the angles of the bounces, and determines the exact geometry and reflectivity of the hidden object. By combining the precise distance measurements from the laser with the predictive power of transient rendering AI, the system generates a clear, 3D reconstruction of the object hidden around the corner.