Multi-Device Groundplot Tracking for Reliable Cashierless Checkout
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional cashierless checkout methods in smart retail stores face challenges with object detection and tracking, particularly due to the reliance on depth cameras or LIDAR systems, which require excessive GPU resources and can result in lost tracking IDs if a camera malfunctions, leading to inappropriate billing and revenue loss.
Innovation Solution
A tracking system that employs multiple media acquisition devices connected via a communication network, allowing for the detection and tracking of objects by projecting input data points onto a groundplot and using a calibration process with matrix multiplication to ensure continuous tracking even if one device fails, with a clustering method to assign and transfer tracking IDs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If groundplot tracking is performed using depth cameras or LIDAR systems attached to the ceiling, then customer detection and tracking is achieved, but excessive GPU resources are required on multiple servers
Solution Approach 1:
The system divides the store into multiple zones with multiple media acquisition devices, each responsible for detecting objects within its specific zone. This segmentation allows distributed processing of detection data, reducing the computational burden on any single server and overall GPU resource consumption while maintaining comprehensive tracking coverage.
Solution Approach 2:
The patent replaces the conventional depth camera or LIDAR-based 3D spatial mapping system with a 2D image processing system using multiple media acquisition devices. By substituting the complex 3D depth measurement mechanism with 2D image capture and groundplot projection, the system significantly reduces computational requirements while achieving equivalent tracking functionality.
2Ease of operation
If tracking ID is assigned based on single depth camera or LIDAR system, then customer tracking is established, but tracking ID may be lost if the device stops operating or malfunctions
Solution Approach 1:
Each media acquisition device is assigned a specific field of view and responsibility zone within the store. The system uses local quality by having each device independently detect and track objects within its local area, with tracking IDs being locally assigned and maintained. This distributed approach ensures that if one device fails, tracking continues uninterrupted by other devices in different zones.
Solution Approach 2:
The system implements redundancy by deploying multiple media acquisition devices across different zones of the store. This beforehand cushioning ensures that if one device malfunctions or stops operating, other devices are already in place to continue tracking objects, preventing loss of tracking ID and maintaining continuous monitoring coverage.
3Area of stationary object
If multiple media acquisition devices are deployed to cover the entire area, then tracking coverage is improved, but system complexity and cost increase
Solution Approach 1:
The system employs multiple media acquisition devices that are identical in function and capability, each serving the same purpose of detecting objects within its zone. This universality simplifies system configuration and deployment, as the same device type can be replicated across multiple locations without requiring complex integration of different device types, thereby reducing overall system complexity despite increased coverage area.
Data Source
AI summary
The embodiments herein relate to surveillance of objects and, more particularly, to efficient detection and tracking of objects. A method disclosed herein includes detecting at least one object in the physical store, on receiving media from a plurality of media acquisition devices positioned in the physical store. The method further includes tracking the at least one object in the physical store by projecting input data points of each media acquisition device onto a groundplot, clustering the input data points into a single cluster, and assigning a tracking identifier (ID) to a centroid of the single cluster, wherein the centroid depicts the at least one object.