Multi-Camera Object Search Using 3D Spatial Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking technologies struggle to recognize objects with dissimilar front and back views across multiple cameras, leading to misidentification as new objects when an object disappears from one camera.
Innovation Solution
A method and apparatus for searching the same object in multi-camera images by using gallery images obtained through multiple cameras, which involves obtaining a query image, comparing it with gallery images, ranking similarities, and recommending images based on these similarities, allowing for repeated iterations with recommended images as new queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If object tracking is performed using conventional methods with single-camera or simple multi-camera systems, then the tracking process is straightforward, but objects with dissimilar front and back views are misidentified as new objects when transitioning between cameras
Solution Approach 1:
The patent transitions from 2D image matching to 3D spatial reasoning by introducing depth information and spatial relationships. The system creates a 3D spatial map of the environment and uses it to predict object positions across multiple cameras, enabling reliable recognition of objects with dissimilar views by considering their spatial trajectories rather than just visual appearance.
Solution Approach 2:
The patent introduces a 3D spatial map as an intermediary between multiple cameras and object tracking algorithms. This spatial map serves as a mediator that integrates information from multiple camera viewpoints, allowing the system to correlate objects across cameras even when their appearances differ significantly, thereby improving recognition reliability without requiring direct complex multi-camera coordination.
2Measurement precision
If multiple cameras are used to capture objects from different directions, then object recognition accuracy improves, but the complexity of comparing and matching images across different viewpoints increases significantly
Solution Approach 1:
The patent changes the parameters used for object identification from appearance-based features (color, shape, texture) to spatial-based parameters (position, trajectory, depth). By transforming the comparison metric from visual similarity to spatial consistency, the system achieves precise object identification across different viewpoints without the complexity of comparing dissimilar visual appearances.
Solution Approach 2:
The patent segments the object tracking problem into independent spatial components by creating a 3D spatial map that divides the environment into discrete depth layers and spatial zones. This segmentation allows the system to track objects independently in different spatial regions and reconstruct their trajectories, simplifying the comparison process across multiple cameras by focusing on spatial continuity rather than visual matching.
3Reliability
If conventional object tracking methods are used, then the system remains simple, but objects disappearing from one camera cannot be reliably found in other cameras
Solution Approach 1:
The patent performs preliminary actions by pre-building a 3D spatial map of the environment and pre-calculating expected object trajectories before objects actually transition between cameras. This preliminary spatial modeling enables the system to predict where objects should appear in subsequent cameras, allowing reliable continuity tracking without requiring complex real-time coordination between multiple cameras.
Data Source
AI summary
Proposed is a method and apparatus for searching for the same object based on multi-camera images, the apparatus and method recognizing objects recorded from different directions through multiple cameras as the same object. An apparatus for searching for the same object based on multi-camera images that is operated by an object search model in accordance with an instruction from a controller is also proposed. A computer-readable recording medium on which a program for performing the method is recorded is also proposed.