Mobile Image Classification With Local KPI Generation Offline
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for object identification and KPI generation in a cloud environment require high-capacity internet connections, leading to costly infrastructure, long processing times, and inefficiencies, especially when dealing with large environments like supermarkets.
Innovation Solution
A local image processing method and system that uses a mobile device to capture images, define operating segments, receive specialized models, and perform object identification and classification without internet connection, optimizing data processing and generating KPIs locally.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If cloud-based image processing is used, then object identification and classification can be performed, but processing time increases and internet connection requirements increase
Solution Approach 1:
The patent segments the image processing task by dividing the image into multiple regions of interest (ROIs) corresponding to different shelves or display areas. Each ROI is processed independently through the neural network, allowing parallel computation and reducing overall processing time while maintaining identification accuracy across the entire image.
Solution Approach 2:
The patent performs preliminary actions by pre-processing the image to identify and extract potential objects before feeding them to the neural network. This includes detecting object candidates, determining their locations, and preparing them for classification, which reduces the computational burden during the main processing phase.
2Measurement precision
If cloud-based image processing is used, then object identification can be performed, but infrastructure costs increase
Solution Approach 1:
The patent implements self-service by enabling the mobile device to perform image processing locally using a neural network model stored in its memory. This eliminates the need for continuous cloud connectivity and expensive infrastructure, allowing the device to independently identify and classify objects without requiring external servers or complex backend systems.
Solution Approach 2:
The patent uses a simplified copy of the neural network model that is optimized for mobile devices. Instead of requiring the full cloud-based processing system, a condensed version of the model is deployed locally on the mobile device, maintaining essential identification capabilities while reducing infrastructure requirements.
3Adaptability or versatility
If global models for all classes of items are used, then comprehensive object identification is possible, but model size becomes excessively large
Solution Approach 1:
The patent applies local quality by training and using specialized neural network models for specific product categories or classes relevant to the retail environment. Instead of using a single global model for all possible objects, the system employs category-specific models that are optimized for identifying products on shelves, thereby reducing model size while maintaining comprehensive coverage for the intended application.
4Area of stationary object
If multiple images are captured for large environments, then complete coverage is achieved, but processing complexity increases
Solution Approach 1:
The patent segments the retail environment into multiple zones or shelves and captures images of each zone separately. The mobile device then processes these segmented images independently using the neural network, and the results are aggregated to provide comprehensive coverage of the entire environment. This segmentation approach reduces processing complexity compared to capturing and processing one large image of the entire space.
Data Source
AI summary
The present invention relates to a local image processing method and system for object identification, classification and generation of at least one KPI based on: capturing an image using a mobile device (20), wherein the image contains at least one specific object type, assigning a specialized model related to the image, wherein the specialized model is related to the specific object type in the image, recognizing at least one object in the image based on the specialized model, informing the user that the said object has been recognized, and calculating at least one KPI related to said object.