Sketched Lidar: Efficient Representation of Time-of-Flight Data

Single-photon light detection and ranging (lidar) has become a prominent tool for depth imaging in recent years. The technique consists of measuring the depth of a target by constructing a histogram of time delays between emitted light pulses and detected photon arrivals. A major data processing bottleneck arises on the lidar device when either the number of photons per pixel is large or the time-stamp resolution is fine, as both the memory and time complexities of the image reconstruction algorithms scale with these parameters. In this talk, we show that it’s possible to circumvent this limiting bottleneck by building a compressed statistic, or a so-called sketch, of the time of flight model that is sufficient to infer the spatial distance and intensity of the objects in the scene. Fundamentally, the size of the sketch scales with the number of surfaces in the scene and is completely independent to the number of photons or the time-stamp resolution of the device. By developing a sketch-based reconstruction algorithm, we show using synthetic and real data that substantial compression can be obtained without sacrificing the quality of the depth images.