Camera Positioning via Deep Learning and Inter-Frame Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current indoor positioning methods, such as vision odometer-based and scene-based end-to-end positioning, face challenges in achieving accurate positioning without additional hardware costs and maintaining robustness against environmental changes.
Innovation Solution
A positioning method utilizing a deep learning model, specifically a PoseNet network, to acquire and update camera coordinates in a world coordinate system by inter-frame matching and landmark detection, eliminating the need for additional hardware like WiFi or Bluetooth modules, and improving accuracy through Kalman filtering and landmark-based positioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If vision odometer-based positioning method is used, then positioning can be performed using only camera, but cumulative error needs to be eliminated every predetermined time and initial coordinate needs to be determined in advance
Solution Approach 1:
The patent applies preliminary action by detecting landmarks in advance to determine the initial coordinate of the camera in the world coordinate system. The system pre-identifies landmark positions and uses them to establish an accurate starting point for positioning, eliminating the need for complex initial coordinate determination procedures and reducing cumulative errors from the outset.
2Ease of operation
If scene-based end-to-end positioning method is used, then positioning is performed directly based on image scene, but positioning accuracy is insufficient
Solution Approach 1:
The patent introduces landmarks as intermediary objects between the camera and the positioning system. Instead of directly determining position from general scene features, the system uses specifically identified landmarks with known world coordinate positions as mediators to calculate camera position, thereby improving accuracy while maintaining operational simplicity.
3Measurement precision
If additional hardware like WiFi or Bluetooth modules is added, then positioning accuracy can be improved, but hardware cost increases
Solution Approach 1:
The patent applies self-service by enabling the camera system to determine its own position using only visual information from images. The system processes image data, detects landmarks, and calculates camera coordinates autonomously without requiring additional positioning hardware, thereby maintaining low hardware complexity while achieving high positioning accuracy.
4Measurement precision
If landmark detection is performed, then initial coordinate accuracy is improved, but processing time increases
Solution Approach 1:
The system performs preliminary landmark detection and stores landmark position information in advance. By pre-identifying landmarks and their world coordinate positions, the system reduces real-time processing requirements while maintaining high initial coordinate accuracy, thereby balancing processing time with positioning precision.
Data Source
AI summary
The embodiments of the present disclosure provide a positioning method, a positioning device. The method may include: acquiring an image from a camera; obtaining a first coordinate of the camera in a world coordinate system from the image based on a deep learning model; obtaining an initial coordinate of the camera in the world coordinate system based on the first coordinate; and determining a real-time coordinate of the camera in the world coordinate system through inter-frame matching based on the initial coordinate of the camera in the world coordinate system.