Traditional computer vision is often limited by a "keyhole" perspective—standard cameras capture only a narrow slice of the world. To map a full room or inspect a tunnel, robots typically have to rotate their sensors or stitch together multiple images, a process that is slow and prone to "ghosting" artifacts where moving objects appear in multiple places.
A robust alternative has emerged: single-shot omnidirectional sensing. By combining specialized mirrors, wide-angle lenses, and structured light, researchers have developed systems that capture a full 360-degree 3D map of an environment in a single instant.
Folding the World with Catadioptric Optics
The most intuitive way to see in all directions at once is to look at a curved mirror. This is the "catadioptric" approach (combining catoptrics, or mirrors, with dioptrics, or lenses).
Early systems like COPIS (Conic Projection Image Sensor), evaluated by Ryota Hiura and Masahiko Yachida in Evaluation of Effectivity of Omnidirectional Image Sensor COPIS in a Real Environment, used a conic mirror to project a 360-degree "bird's-eye" view onto a single camera sensor. Because vertical lines in the real world (like doorframes) appear as radial lines in this projection, the system can triangulate the robot's position and detect obstacles by simply tracking how these radial lines move.
To gain depth information without moving the camera, researchers like Trung Ngo Thanh and Masahiko Yachida developed multi-mirror sensors. As described in Real-Time Estimation of Fast Egomotion with Feature Classification Using Compound Omnidirectional Vision Sensor, these devices use a central mirror for a wide field of view and peripheral mirrors to provide "multi-baseline" stereo. This allows the system to calculate depth in a single frame, enabling robots to estimate their own motion even when moving at high speeds.
Active Light Sectioning in 360 Degrees
While mirrors capture ambient light, "active" systems project their own light to measure geometry. In confined spaces like pipes or tunnels, where textures are often uniform and featureless, standard cameras struggle.
Atsushi Yamashita pioneered a technique called omnidirectional active light-sectioning. By pairing a 360-degree camera with a ring-shaped laser, as detailed in 3-D Shape Reconstruction of Pipe with Omnidirectional Laser and Omnidirectional Camera, the system projects a continuous plane of light onto the interior walls. The camera captures the entire laser ring in one shot; any deformation in the ring's shape in the image directly corresponds to the 3D geometry of the pipe.
This concept was further refined for larger environments like tunnels by Takuya Igaue and Atsushi Yamashita in Line Structured Light-Based Tunnel 3D Measurement with 2D-3D Point Cloud Matching on Projected Light of Ring Laser. By using the laser line itself as a reference for matching, they can stitch together dense point clouds even in smooth, featureless environments.
One-Shot Depth and Semantic Mapping
The modern frontier of this research moves beyond simple geometry to full semantic understanding. The goal is to go from a single 360-degree image to a complete, labeled top-down map (a "Bird's-Eye View" or BEV).
Zhifeng Teng and Kailun Yang introduced 360Mapper in 360BEV: Panoramic Semantic Mapping for Indoor Bird's-Eye View, which uses depth-aware projection to transform a single panorama into a semantic map. This bypasses the need for complex multi-camera calibration.
To support these AI models, researchers are developing "one-shot" hardware that captures both color and depth simultaneously. Hideaki Kanayama and Didier Stricker introduced the ToF-360 dataset in ToF-360 - A Panoramic Time-of-flight RGB-D Dataset for Single Capture Indoor Semantic 3D Reconstruction. By using integrated Time-of-Flight (ToF) sensors and fisheye cameras, they can capture a full 3D scene in a single second, eliminating the geometric misalignments that plague traditional scanning.
Perhaps the most extreme version of this "static sensing" is the work of Mohit Gupta and Shree K. Nayar. In Trapping Light for Time of Flight, they describe a "light trap"—a mirror-filled box that allows a single ToF sensor to see an object from all sides simultaneously. By computationally "unrolling" the reflected light paths, they can reconstruct a complete 360-degree 3D model of an object without ever moving the camera or the object itself.