Local Object Detection Using Remotely Generated Specialized Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image recognition systems for monitoring large areas, such as stores, face challenges due to limited memory on single board computing devices, leading to inadequate detection performance and high costs from relying on cloud computing and manual configuration of models on numerous devices.
Innovation Solution
Automated techniques for quickly and easily installing a different image recognition model on each device by capturing data, transmitting it to a remote system for processing using various machine learning techniques, and generating configuration data for local installation, enabling efficient object detection and tracking without manual configuration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a generic machine learning model is installed on every device to cover a whole store or department, then the detection coverage is improved, but the memory resources on single board computing devices are insufficient
Solution Approach 1:
The patent divides the generic machine learning model into multiple specialized models, each trained to detect specific product categories or subsets. Each single board computing device loads only its assigned specialized model rather than a complete generic model, reducing memory requirements while maintaining overall store-wide detection coverage through coordinated operation of multiple devices
Solution Approach 2:
The patent assigns different specialized models to different devices based on their specific monitoring zones and product categories. Each device is optimized with a model tailored to its local context, improving detection accuracy for that specific area while using minimal memory resources appropriate for the device's constrained capabilities
2Quantity of substance
If a small generic model is installed on each device to fit limited memory, then the memory usage is reduced, but the detection performance and accuracy are inadequate
Solution Approach 1:
Instead of compressing a small generic model, the patent segments the detection task into multiple specialized models, each focused on specific product categories. This allows each model to be highly specialized and accurate for its domain while remaining small enough for single board computing devices, achieving both low memory usage and high detection accuracy
Solution Approach 2:
The system automatically generates and assigns specialized models to devices based on their monitoring zones and product catalogs, eliminating the need for manual configuration. Each device receives a model optimized for its specific context, ensuring high detection accuracy without requiring large memory resources
3Power
If cloud computing resources are used to perform image recognition, then the computational power is improved, but the costs and network traffic increase
Solution Approach 1:
The patent extracts the machine learning model execution from cloud computing resources and places specialized models directly on single board computing devices at store locations. This enables local processing of images captured by depth cameras, eliminating the need to transmit images to the cloud and reducing both cloud computing costs and network traffic while maintaining sufficient computational power for object detection
Solution Approach 2:
Each device performs its own image processing locally using the specialized model installed on it, rather than relying on external cloud resources. The device autonomously detects objects in its monitoring zone, generating results without requiring cloud computing power, thereby reducing operational costs and network dependency
4Measurement precision
If manual configuration of different models on each device is performed, then the detection performance is improved, but the time and cost increase significantly
Solution Approach 1:
The system automatically generates specialized models for each device based on its monitoring zone, product catalog, and detected objects, eliminating the need for manual configuration. The automated process assigns appropriate models to devices and configures them to detect specific product categories, achieving high detection performance while reducing configuration time and costs
Solution Approach 2:
The system pre-trains specialized models for different product categories and automatically selects and assigns the appropriate models to devices based on their specific contexts before deployment. This preliminary preparation of models and automatic assignment process eliminates the need for time-consuming manual configuration while ensuring each device has the optimal model for its detection tasks
Data Source
AI summary
Embodiments of the present invention may provide automated techniques for quickly and easily installing a different model for image recognition on each of numerous devices without doing manual configuration of each and every device. For example, in an embodiment, a computer-implemented method for configuring devices may comprise capturing data at a device, transmitting the captured data to a remote system, receiving configuration data for the device, wherein the configuration data has been generated by processing the captured data at the remote system using a plurality of different machine learning techniques and generating the configuration data based on at least one of the plurality of different machine learning techniques, and configuring the device using the received configuration data.