Multi-Camera Human Tracking with Dynamic Similarity Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for monitoring and analyzing human figures in stores using multiple cameras face challenges in accurately comparing face images due to variations in illumination and camera orientations, leading to errors in movement path and stay time analysis.
Innovation Solution
An information processing apparatus with multiple units for image acquisition, detection, tracking, and feature extraction, which searches for objects with similarity thresholds to accurately identify and track human figures across multiple cameras, reducing errors and omissions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If face images are compared using a single threshold for similarity, then the comparison process is simple, but search omissions occur when the threshold is increased or false matches occur when the threshold is lowered
Solution Approach 1:
The patent applies dynamic threshold adjustment by performing face image comparison at multiple different threshold values rather than using a single fixed threshold. The system dynamically selects appropriate thresholds based on the specific comparison context, allowing it to balance between avoiding search omissions and preventing false matches, thereby improving search accuracy without requiring overly complex fixed-threshold designs
Solution Approach 2:
The patent changes the similarity threshold parameter during the comparison process. By comparing face images using multiple different threshold values and selecting results based on these varying parameters, the system achieves more reliable search results while maintaining reasonable process complexity
2Quantity of substance
If face images are additionally registered sequentially to handle comparison failures, then more face images are available for comparison, but comparison errors may still occur and the process becomes more complex
Solution Approach 1:
The patent performs preliminary face image registration by storing multiple face images of the same person in advance before the actual search operation. This preliminary action ensures that when a face image comparison is needed, there are multiple registered images to compare against, reducing the likelihood of comparison failures while maintaining systematic process control
Solution Approach 2:
The patent implements a feedback mechanism where the system evaluates comparison results and uses this information to guide subsequent comparison operations. By analyzing whether comparisons succeeded or failed and adjusting the selection of registered face images accordingly, the system improves comparison accuracy without requiring uncontrolled sequential registration of all possible face images
3Device complexity
If a single face image is used for comparison, then the registration process is simple, but illumination and orientation variations cause similarity deterioration
Solution Approach 1:
The patent segments the face identification process by using multiple different registered face images of the same person rather than relying on a single image. This segmentation approach allows the system to handle variations in illumination and orientation by having multiple specialized images available for comparison, improving measurement precision while keeping the registration process manageable
Solution Approach 2:
The patent creates a composite reference by combining multiple face images of the same person into a registered set. This composite approach, where multiple images work together as a unified reference system, enables the system to overcome limitations of individual images affected by illumination and orientation variations, thereby improving similarity measurement accuracy
Data Source
AI summary
An apparatus which detects an object in a frame image of a moving image captured by each image capturing apparatus in an area, acquires an image feature of the detected object, tracks the detected object in one moving image, acquires an image feature of an object to be searched from the detected object, searches for an object having a similarity equal to or higher than a predetermined threshold from the acquired image features and identifies the object, acquires an image feature of the tracked object and corresponding to the identified object from the acquired image feature, and searches for an object having a similarity equal to or higher than the predetermined threshold from the acquired image features and identifies the object.