Vehicle Localization Using Surveillance Cameras and Odometry Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for determining the position of personal-mobility vehicles (PMVs) and autonomous mobile robots (AMRs) in environments like airports, hospitals, and shopping malls are imprecise and costly, requiring additional infrastructure and calibration, especially in indoor settings where satellite navigation is unavailable.
Innovation Solution
Utilizing surveillance cameras installed in the environment to detect vehicles with unique visual patterns, processing images to associate floor coordinates with map coordinates, and determining the odometry center of the vehicle through pattern recognition, enabling precise positioning without additional infrastructure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If additional infrastructure (mobile communication radio transmitters, visual feature learning) is installed for positioning, then positioning precision is improved, but device complexity and installation cost increase
Solution Approach 1:
The patent reuses existing surveillance cameras for their primary security function while adding a secondary positioning function. The same camera infrastructure serves both security monitoring and vehicle/robot positioning, eliminating the need for dedicated positioning infrastructure.
Solution Approach 2:
The system uses the environment's existing surveillance infrastructure to provide positioning services without requiring the environment to be modified or equipped with additional positioning-specific devices. The surveillance system serves itself by providing dual functionality.
2Measurement precision
If additional infrastructure (mobile communication radio transmitters, visual feature learning) is installed for positioning, then positioning precision is improved, but installation cost increases
Solution Approach 1:
The patent reuses existing surveillance cameras for their primary security function while adding a secondary positioning function. The same camera infrastructure serves both security monitoring and vehicle/robot positioning, eliminating the need for dedicated positioning infrastructure.
Solution Approach 2:
The system uses inexpensive visual patterns (stickers, markers) that can be easily applied and removed from vehicles/robots, replacing expensive dedicated positioning hardware. These visual features are simple, cheap, and sufficient for the positioning task.
3Device complexity
If odometry is used for position estimation, then positioning is achieved without additional infrastructure, but measurement precision deteriorates due to error accumulation
Solution Approach 1:
The system uses periodic visual recognition by surveillance cameras to correct and reset odometry accumulation errors. The camera-based absolute position measurements provide feedback that recalibrates the odometry system, maintaining long-term positioning accuracy without requiring additional infrastructure.
Solution Approach 2:
The patent combines odometry (for continuous, infrastructure-free tracking) with periodic visual recognition from existing surveillance cameras (for absolute position correction). This hybrid approach leverages the strengths of both methods while mitigating their individual weaknesses.
Data Source
AI summary
Described herein are solutions for locating a vehicle (1a) in an environment through a plurality of surveillance cameras (2) installed in the environment. The vehicle (1a) comprises a plurality of sensors (50) configured to detect data (S2) that identify a displacement of the vehicle (1a), and the vehicle (1a) estimates a position (POS′) of an odometry centre (OC) of the vehicle (1a) via odometry as a function of the data (S2) that identify the displacement of the vehicle (1a). A plurality of visual patterns (P) are applied to the vehicle (1a). During a learning phase (1100), a processor (3a) receives a map (300) of the environment and, for each camera (2), an image (306) acquired by the respective camera (2). Next, the processor (3a) generates data (308) that enable association of a pixel of a floor/ground (310) in the image (306) to respective co-ordinates in the map (300). During a localization phase (1200), the processor (3a) repeats (1206, 1210) a sequence of steps for at least one of the surveillance cameras (2). In particular, the processor (3a) receives (1250) an obfuscated image (312) from the camera (2) and checks whether the obfuscated image (312) presents one or more of the visual patterns (P) applied to the vehicle (1a). In the case where the obfuscated image (312) presents one or more of the visual patterns (P), the processor (3a) calculates (1252) the position of an odometry centre (OC) in the obfuscated image (312) as a function of the positions and optionally of the dimensions of the visual patterns (P) appearing in the obfuscated image (312). Next, the processor (3a) determines (1254) a position (POS) of the odometry centre (OC) in the map (300) by mapping the position in the obfuscated image (312) into coordinates in the map (300), using the data (308) that enable association of a pixel of a floor/ground (310) in the image (306) to respective co-ordinates in the map (300). Finally, the processor (3a) sends (1212) the position (POS) of the odometry centre (OC) in the map (300) to the vehicle (a), and the vehicle (1a) sets the estimated position (POS′) of the odometry centre (OC) at the position (POS) received.