Adaptive Epipolar Time-of-Flight Imaging

Autonomous robots rely on 3D sensors to safely navigate and interact with their environment. The most reliable sensing solution is light detection and ranging (LIDAR) which uses the time-of-flight (ToF) principle to measure the distance to scene points. These long-range systems typically scan one or more lasers across the scene but the scanning is slow and although speed can be increased with more lasers, this further increases the complexity, size, weight, power, and cost (SWaP-C) of these already expensive systems. Additionally, even high-end LIDAR sensors do not have enough spatial resolution to both detect small objects at long ranges and image a wide FoV. Alternatively, flash LIDAR uses a powerful broad flash of light to capture the scene in a single snapshot. The resulting increase in speed, however, comes at a large cost in light efficiency since the system’s light power is now spread out across the entire scene. A fundamental limitation of these LIDAR systems is that they have few adjustable parameters and cannot adapt to the sensing needs of a particular situation. The proposed approach develops an adaptive ToF imaging method to provide a robust, fast, and efficient 3D sensing solution. First, it utilizes a powerful 1D scanning approach innovated by the proposing team called epipolar imaging to improve efficiency. This method sweeps a line of sensing from the rolling shutter of a 2D image sensor across the scene in sync with an aligned plane of illumination. Due to the concentration of the available light into a single line of sensing, this 1D scanning method is more efficient and provides longer ranges than global shutter systems at speeds faster than point scanning systems. To further improve the efficiency of epipolar imaging, this work will extend the approach with an adaptive illumination system that can redistribute and steer available photons along the plane to illuminate only as much of the scene that is needed. Guided by adaptive sampling algorithms developed in the proposed work, this approach will enable an even more efficient system through the dynamic allocation of light power to key areas of the scene. The proposed research and development effort first applies this method to continuous-wave time-of-flight (CW-ToF) sensors to build a high-resolution, robust, and low SWaP-C 3D sensor for mid-range applications in object detection, tracking, and recognition. Phase I research will include the development, integration, and demonstration of the high-resolution epipolar ToF imaging system with the adaptive illumination source. The base effort will first focus on hardware and system development. The following option effort will develop and demonstrate the algorithms used to adaptively sample a scene for improved object detection and tracking in a variety of real-world situations.