Point Cloud Alignment for Indoor Camera Localization

The patent creates a digital copy of the physical environment through point cloud representations. A reference point cloud is constructed from high-precision LiDAR scans of the environment, storing geometric information without requiring the physical LiDAR device during operation. This allows depth measurement functionality to be replicated through computational methods rather than requiring expensive hardware copies.

The patent replaces the mechanical/optical LiDAR measurement system with a computational approach using standard camera sensors. Instead of using time-of-flight or laser ranging hardware, the system uses photogrammetry algorithms to extract depth information from sequential images, substituting mechanical measurement systems with software-based computational methods.

The patent replaces expensive, specialized LiDAR sensors with inexpensive standard camera modules that are already widely available in mobile devices. The system uses multiple low-cost image sensors rather than one high-cost depth sensor, achieving comparable or superior functionality through redundancy of cheap components rather than a single expensive component.

The patent makes the camera system multi-functional by enabling it to perform both standard 2D imaging and 3D depth measurement tasks. The same image sensor and processing pipeline used for photography are also used for photogrammetric depth reconstruction, eliminating the need for separate specialized depth-sensing hardware and reducing overall device cost.

The patent transitions from 2D image data to 3D spatial information by constructing point clouds that represent the third dimension (depth). By processing sequential 2D images through photogrammetry algorithms, the system recovers 3D geometric structure, enabling absolute depth measurements without adding depth-sensing hardware dimensions.

The patent performs preliminary high-precision scanning of the environment to create a reference point cloud before operational use. This pre-captured geometric information serves as a reference framework that enables subsequent real-time measurements to achieve high precision without requiring the expensive scanning hardware during actual operation, separating the precision measurement phase from the operational phase.

The patent performs computationally intensive photogrammetry processing and point cloud construction in advance, before real-time operation begins. The reference point cloud is pre-processed and stored, so that during actual use, the system only needs to perform lighter computational tasks like comparing current frame data against the pre-built reference, dramatically reducing real-time processing requirements.

The patent divides the computational workload into separate stages: offline preprocessing of reference data using heavy photogrammetry algorithms, and online real-time processing using lighter comparison and matching algorithms. This segmentation allows complex absolute depth measurement computations to be performed when processing time is not constrained, while real-time operation maintains low latency.