Local Object Identification for Image Data Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Personal devices equipped with image sensors face limitations in processing and storage, restricting the variety of objects that can be identified and the information available about them, necessitating offloading object identification operations to search and analysis systems, which can be resource-intensive and delayed due to network latency and bandwidth constraints.
Innovation Solution
The implementation of an image capturing device that uses local object identification techniques to determine object types and characteristics, such as barcodes, faces, and landmarks, and selectively offloads requests to a search and analysis system based on resource usage and focus distance, reducing unnecessary data transfer and improving processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If object identification operations are offloaded to a search and analysis system, then the variety of objects that can be identified increases, but network latency and bandwidth constraints cause delays and increased resource consumption
Solution Approach 1:
The patent segments object identification into two parts: common objects (barcodes, faces, landmarks) are identified locally using lightweight machine learning models, while more complex or unknown objects are offloaded to the search and analysis system. This segmentation allows immediate identification of frequent objects without network latency, while still providing the capability to identify a wide variety of objects through selective offloading.
2Adaptability or versatility
If object identification operations are offloaded to a search and analysis system, then the variety of objects that can be identified increases, but resource consumption and network bandwidth usage increase
Solution Approach 1:
The system segments data transfer by only sending images to the search and analysis system when local identification fails or when the object type is unknown. This selective offloading reduces network bandwidth consumption and energy usage while maintaining the ability to identify a wide variety of objects through the search and analysis system when needed.
3Productivity
If local object identification is performed for all object types, then processing speed increases, but processing and storage limitations restrict the variety of objects that can be identified
Solution Approach 1:
The system performs preliminary local identification using lightweight machine learning models for common object types (barcodes, faces, landmarks). This preliminary action quickly identifies the majority of objects without network latency. When the local model cannot identify an object or the object type is unknown, the system then performs the secondary action of offloading to the search and analysis system, ensuring both speed and versatility.
Data Source
AI summary
A method, a device, and a computer program product for identifying objects in image data. The device is enabled to, in response to determining that an object type of a first set of object types is represented in the image data, indicate that the object type is represented in captured image data. The device is further enabled to, in response to determining that no object type of the first set of object types is represented in the image data, determine whether the image data comprises an object characteristic of a specific one of a second set of object types. The device is further enabled to, in response to determining that the image data comprises an object characteristic of the specific one of the second set of object types, indicate that the image data comprises an object characteristic of the specific one of the second set of object types.