Adaptive Object Recognition on Mobile Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computing devices, such as smartphones, have limited ability to recognize objects within digital images stored on them, necessitating a technical process to enhance object recognition capabilities.
Innovation Solution
An adaptive object recognition system that selectively loads and updates object recognition definitions based on available resources, using a compressed object recognition model to analyze digital images and transmit metadata for recognized objects, leveraging machine learning techniques and network resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the entire object recognition model is loaded into memory, then object recognition capability is improved, but memory resource consumption increases
Solution Approach 1:
The patent divides the object recognition model into multiple independent object recognition definitions, each representing a separate object class or category. These segmented definitions can be loaded independently into memory based on available resources and specific recognition needs, rather than loading the entire model at once.
Solution Approach 2:
The system dynamically loads and unloads object recognition definitions based on real-time memory availability and recognition requirements. The model manager monitors resource conditions and adaptively manages which definitions are resident in memory, enabling the system to maintain adequate recognition capability while adapting to changing resource constraints.
2Adaptability or versatility
If multiple object recognition definitions are loaded simultaneously, then recognition versatility is improved, but memory usage increases
Solution Approach 1:
The system design allows a single object recognition model to support multiple object recognition definitions that can be selectively activated. This universal architecture enables the same base model to recognize different types of objects by loading only the relevant definitions needed for current tasks, providing versatility without requiring all definitions to be permanently loaded.
Solution Approach 2:
The system changes the operational parameters by dynamically selecting which object recognition definitions are active based on current needs. Instead of maintaining a fixed set of loaded definitions, the system adjusts the configuration of active definitions according to available memory and specific recognition tasks, optimizing the balance between versatility and resource usage.
3Measurement precision
If object recognition definitions are updated frequently, then recognition accuracy is improved, but processing time increases
Solution Approach 1:
The system performs preliminary actions by pre-processing and validating object recognition definitions before they are loaded into memory. The model manager prepares definitions in advance, checking their compatibility and readiness, so that when loading is required, the process is faster and more efficient, reducing the time penalty associated with updates.
Solution Approach 2:
The system implements feedback mechanisms where the model manager monitors the performance and resource usage of loaded definitions. Based on this feedback, the system intelligently determines when updates are truly necessary versus when existing definitions remain adequate, reducing unnecessary update operations and their associated processing time costs while maintaining recognition accuracy.
Data Source
AI summary
A digital image processing method including: retrieving by a mobile computing device an object recognition model, where the object recognition model includes one or more object recognition definitions usable to identify within a digital image one or more objects captured by the digital image; receiving a first indication from the mobile computing device that there is a first computer resource available; determining, based on the first computer resource available, that the first computer resource available is sufficient to load a first object recognition definition defined in the object recognition model; loading into the first computer resource the first object recognition definition; analyzing a particular digital image, using the first object recognition definition loaded into the first computer resource, to determine whether a first object defined by the first object recognition definition is captured within the particular digital image; receiving a second indication from the mobile computing device that there is a second computer resource available; determining, based on the second computer resource available, that the second computer resource available is sufficient to load a second object recognition definition defined in the object recognition model; loading into the second computer resource the second object recognition definition; analyzing the particular digital image, using the second object recognition definition loaded into the second computer resource, to determine whether a second object defined by the second object recognition definition is captured within the particular digital image; associating as metadata of the particular digital image that the first object or the second object has been captured within the particular digital image.