Edge Device Object Recognition Database Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
IoT edge devices with limited computing resources and bandwidth face challenges in establishing high-speed network connections and performing efficient object recognition due to large reference image databases, which require significant processing power and power consumption.
Innovation Solution
A system comprising a server computing device that maintains a reduced object recognition database for each edge computing device, using three-dimensional models to generate two-dimensional images, and only performing object recognition in response to device events, thereby reducing processing and power requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large reference image database is used for object recognition, then object recognition accuracy is improved, but processing power requirements and power consumption increase significantly
Solution Approach 1:
The patent segments the object recognition database into multiple specialized databases organized by object categories (e.g., animals, vehicles, plants). Each database contains only images relevant to its category, allowing the system to select and use only the appropriate database for a given recognition task rather than processing through a single large database containing all objects.
Solution Approach 2:
The system performs partial action by selecting and using only the subset of databases relevant to the current recognition task. Instead of processing through the entire large database, the system activates only the necessary category-specific databases, reducing computational effort while maintaining recognition accuracy for the target objects.
2Measurement precision
If a large reference image database is used for object recognition, then object recognition accuracy is improved, but processing speed decreases due to increased execution time
Solution Approach 1:
The patent segments the object recognition database into multiple specialized databases organized by object categories (e.g., animals, vehicles, plants). Each database contains only images relevant to its category, allowing the system to select and use only the appropriate database for a given recognition task rather than processing through a single large database containing all objects.
Solution Approach 2:
The system performs partial action by selecting and using only the subset of databases relevant to the current recognition task. Instead of processing through the entire large database, the system activates only the necessary category-specific databases, reducing computational effort while maintaining recognition accuracy for the target objects.
3Measurement precision
If full object recognition processing is performed continuously, then object identification accuracy is maintained, but processing power requirements and power consumption increase
Solution Approach 1:
The patent implements periodic action by triggering full object recognition processing only in response to specific device events (e.g., detected motion, door opening, scheduled intervals) rather than continuously. Between events, the system remains in a lower-power state, significantly reducing overall power consumption while maintaining the ability to accurately identify objects when needed.
4Adaptability or versatility
If a complete object recognition database is deployed to each edge device, then object recognition capability is comprehensive, but device storage and processing requirements increase
Solution Approach 1:
The patent segments the object recognition database into multiple specialized databases organized by object categories (e.g., animals, vehicles, plants). Each database contains only images relevant to its category, allowing the system to select and use only the appropriate database for a given recognition task rather than processing through a single large database containing all objects.
Solution Approach 2:
The system applies local quality by tailoring the database content to match the specific needs and environment of each edge device. Different devices receive different subsets of category databases based on their location, function, and expected recognition tasks, optimizing both storage utilization and recognition performance for local conditions.
Data Source
AI summary
Technologies for edge device object recognition include a server and one or more edge devices in communication over a network. The server maintains an object recognition database that stores images associated with a plurality of objects. The server identifies a subset of the objects that is expected to be recognized by each of the edge devices and generates a reduced object recognition database for each edge device that includes the corresponding subset of images. Each edge device monitors for device events and, in response to a device event, performs object recognition using the corresponding reduced object recognition database. The edge device may transmit thumbnail images of unrecognized objects to the server. The edge device may be coupled to a product storage device such as a cooler or retail shelf, and device events may include sensor events such as door open or door closed events. Other embodiments are described and claimed.