Edge-Based Best-Image Selection for Retail Security Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large retail environments, in-store employees struggle to track and associate guests with potential security threats or theft due to the scale of the environment and the need to manually monitor multiple camera feeds, often missing suspicious activities and relying on subjective assessments.
Innovation Solution
A system using edge computing devices with object detection and machine learning techniques generates and selects best digital images of guests from a continuous video stream, leveraging low computational resources to efficiently isolate and score facial features, allowing for objective association with security events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If security camera systems capture continuous footage of guests in large retail environments, then security monitoring capability is improved, but the complexity of manually reviewing and associating footage across multiple camera feeds increases significantly
Solution Approach 1:
The patent introduces an intermediary system comprising object detection models and image selection algorithms that automatically process camera feeds. This intermediary layer extracts key information (guest detections with bounding boxes) and selects representative images, mediating between the raw camera footage and human reviewers. This resolves the contradiction by maintaining security monitoring reliability while eliminating the complexity of manual multi-feed review through automated image selection based on detection confidence and temporal spacing.
2Reliability
If in-store employees manually monitor multiple camera feeds to track guests, then security event detection is improved, but the time required for manual review and association increases
Solution Approach 1:
The system performs preliminary actions by automatically detecting objects in camera feeds, generating bounding boxes, and pre-selecting representative images before human review. The object detection models continuously process footage and identify potential security events, preparing curated image sets in advance. This preliminary automation maintains reliable security event detection while dramatically reducing the time employees spend on manual review, as they receive pre-processed, relevant images rather than reviewing entire camera feeds.
3Measurement precision
If advanced image analysis techniques are used to select best digital images, then image selection accuracy is improved, but computational resource requirements increase
Solution Approach 1:
The patent applies partial action by using object detection models to identify and process only relevant portions of camera feeds—specifically regions containing detected guests with bounding boxes. Rather than analyzing entire frames or all camera feeds equally, the system focuses computational resources on extracting and selecting images of detected objects. This approach maintains high image selection accuracy while reducing computational requirements by processing only the necessary portions of visual data.
Data Source
AI summary
The disclosed technology provides for generating best images of a person in a retail environment. A method may include receiving, by an edge computing device from a camera, a continuous stream of image data of the retail environment, detecting, using object detection techniques, a person in the image data, the image data including a group of images that are part of a time series, generating bounding boxes for each of the group of images around the person based on detecting the person as they move in the images, identifying, based on applying a features model to each bounding box, at least one feature of the group of images depicting the person, selecting a subset of the bounding boxes having at least one feature that satisfies best images criteria, the subset having best images of the person, and returning the best images of the person.