Image Processing Device for Multi-Camera Object Identity Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for identifying identical objects in images from multiple cameras struggle with changes in features caused by background or occlusion, leading to reduced accuracy in determining object identity.
Innovation Solution
An image processing device with units for object detection, tracking, feature descriptor generation, database formation, retrieval, feature extraction, and clustering, which analyzes and compares features across different camera regions to determine object identity, thereby mitigating the effects of background and occlusion changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If features are averaged in the sequence to remove changes caused by background or occlusion, then robustness of feature extraction is improved, but effective features for determining object identity are ignored
Solution Approach 1:
The patent segments the sequence of image data into multiple clusters based on feature similarity, rather than treating the entire sequence as a single averaged feature. This segmentation allows different temporal segments to be analyzed separately, preserving distinctive features in each segment while still achieving robustness through clustering. The clustering process divides the sequence into meaningful groups that maintain characteristic features.
Solution Approach 2:
The patent applies local quality by extracting features from specific clustered segments rather than uniformly averaging the entire sequence. Each cluster represents a local temporal region with its own characteristic features, allowing the system to preserve locally distinctive features (such as occlusion patterns or background changes) that are relevant to object identification while maintaining overall robustness.
2Device complexity
If conventional feature extraction methods are used on still images, then simplicity of processing is maintained, but accuracy in identifying identical objects from moving images is reduced
Solution Approach 1:
The patent performs preliminary clustering of image data sequences before final feature extraction and comparison. By pre-grouping similar temporal segments into clusters, the system prepares the data in a structured format that enhances identification accuracy. This preliminary organization of data allows subsequent comparison operations to be more effective without significantly increasing overall system complexity.
Solution Approach 2:
The patent transitions from analyzing single still images to analyzing sequences of images by adding the temporal dimension. The clustering process organizes this temporal data into meaningful groups, effectively transforming the problem from 2D spatial feature extraction to 3D spatio-temporal feature analysis, thereby improving identification accuracy for moving objects.
Data Source
AI summary
An image processing device (100) includes an object detecting unit (103a) that detects objects each appearing in a corresponding one of pieces of image data; an object tracking unit (103b) that tracks the detected objects in a time direction; a descriptor generating unit (104) that generates feature descriptors of the respective objects on the basis of results of the detection and results of the tracking of the objects; a data recording control unit (105) that forms a database from the generated feature descriptors and the pieces of image data of the objects corresponding to the feature descriptors; an image retrieving unit (108a) that retrieves, from the database, object groups that match a retrieval condition and form a pair of object groups appearing in pieces of image data obtained by capturing different regions, each of the object groups corresponding to a corresponding one of the regions; a feature extracting unit (108b) that extracts a feature to be used for checking, from pieces of image data and feature descriptors of each of the retrieved object groups; and a determining unit (108c) that determines whether the object groups forming the pair of object groups are identical to each other, by checking the extracted features against each other.