Computer Vision Item Search Using Composite Visual Depth Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer vision systems fail to effectively combine visual and depth data to provide contextual information about objects, limiting their application in areas like item search, particularly for clothing and personal items, where size and style are crucial.
Innovation Solution
A system and method that utilize both visual and depth data from cameras to capture and analyze contextual information such as textures, shapes, and sizes of clothing articles, integrating this data with user profiles to facilitate targeted item searches and deliveries, using a combination of local and remote devices for processing and analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If visual data and depth data are provided separately to consuming applications, then device complexity is reduced and data transmission is simplified, but the ability to provide contextual information about objects is limited
Solution Approach 1:
The patent combines visual data and depth data into a composite image structure that preserves both data types while enabling their integrated processing. This merging allows the system to extract contextual information about objects (such as clothing items) by analyzing both appearance and depth characteristics together, resolving the contradiction between data completeness and transmission simplicity.
Solution Approach 2:
The patent segments the composite image into distinct visual data components and depth data components, allowing each to be processed independently where appropriate while maintaining their relationship. This segmentation enables efficient extraction of contextual information without requiring complete re-transmission of all original data.
2Measurement precision
If computer vision systems use only visual data, then processing is simpler and faster, but the ability to accurately identify size and shape of objects is insufficient
Solution Approach 1:
The patent performs preliminary processing of depth data to extract geometric characteristics (such as volume, surface area, and shape descriptors) before combining them with visual data analysis. This preliminary action enables accurate size and shape identification while maintaining processing efficiency by pre-computing depth-based measurements that can be quickly integrated with visual features.
Solution Approach 2:
The patent creates a composite data structure that integrates visual features (color, texture, pattern) with depth-derived geometric properties to form a comprehensive object representation. This composite approach enables precise measurement of size and shape while maintaining processing productivity through optimized data structures and efficient algorithms.
3Adaptability or versatility
If the system processes and analyzes data using local devices only, then data privacy is improved and processing speed is faster, but the capability to provide sophisticated analysis and search functions is limited
Solution Approach 1:
The patent introduces a cloud-based processing service as an intermediary that receives composite image data, performs sophisticated analysis (such as object recognition, attribute extraction, and similarity matching), and returns results to local devices. This intermediary approach enables sophisticated analysis capabilities while minimizing local processing time through distributed computation.
Solution Approach 2:
The patent implements partial processing at local devices (such as basic image capture and preliminary filtering) while delegating more computationally intensive tasks to cloud services. This partial action approach maintains data privacy and speed for simple operations while leveraging cloud resources for sophisticated analysis, optimizing the balance between capability and time.
Data Source
AI summary
System and techniques for computer vision assisted item search are described herein. A composite image, including visual data and depth data, may be obtained. The composite image may be filtered to isolate a clothing article represented in the composite image. A classifier may be applied to the depth data to produce a set of clothing attributes for the clothing article. The clothing attributes may then be provided to a remote device.