Image Processing Device Selective Person Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In similar image search and detection systems, the processing of continuous video streams often results in image or person omission, leading to poor search and detection accuracy due to the inability to handle multiple detections efficiently during registration and collation processing.
Innovation Solution
An image processing device that detects person areas, calculates feature amounts, and selectively prioritizes individuals based on largest image size and minimum past cumulative processing, using random number generation to ensure thorough processing without omitting critical frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If exhaustive processing of all detected persons is performed, then search and detection accuracy is improved, but image or person omission occurs due to processing time constraints
Solution Approach 1:
The patent changes the parameter of person selection from random or sequential to based on image size and cumulative processing count. By selecting persons with larger image sizes (indicating closer proximity and higher detection reliability) and balancing the cumulative processing count, the system optimizes processing efficiency while maintaining high detection accuracy without omitting critical frames.
Solution Approach 2:
The patent applies different processing priorities to different persons based on their local characteristics - specifically image size and processing history. Persons with larger image sizes receive higher priority for processing, and the system balances the cumulative processing count across different persons to ensure thorough processing without excessive time consumption.
2Reliability
If multiple person detections are processed simultaneously, then detection thoroughness is improved, but processing efficiency deteriorates leading to frame omission
Solution Approach 1:
The patent introduces a cumulative processing count parameter to track how many times each person has been processed. By balancing this parameter across different persons, the system ensures that multiple detections are handled thoroughly while maintaining processing efficiency and preventing frame omission.
Solution Approach 2:
The patent dynamically adjusts processing priorities based on real-time parameters including current image size and cumulative processing count. This dynamic selection approach allows the system to adaptively balance thoroughness and efficiency, ensuring that persons detected multiple times are processed appropriately without causing frame omission.
3Measurement precision
If persons with larger image sizes are prioritized for processing, then detection accuracy is improved, but processing balance deteriorates causing some persons to be overlooked
Solution Approach 1:
The patent uses two parameters for person selection: image size (for accuracy) and cumulative processing count (for balance). This dual-parameter approach ensures that persons with larger image sizes are prioritized for accurate detection while the cumulative count mechanism prevents any person from being overlooked, maintaining processing balance across all detections.
Solution Approach 2:
The patent implements feedback through the cumulative processing count mechanism. The system monitors how many times each person has been processed and uses this information to adjust selection priorities, ensuring that persons who have been processed less frequently are given opportunities for processing, thereby maintaining balance while preserving accuracy.
Data Source
AI summary
An image processing device or a method of processing an image is disclosed. The method includes receiving an image, detecting a plurality of person images in the image, identifying at least a first person image from among the plurality of person images by a preferential method, identifying at least a second person image from among the plurality of person images excluding the first person image, by an exhaustive method, and extracting a first feature amount as to the first person image and a second feature amount as to the second person image during a time interval. For example, the total number of the first person images and the second person images is no more than a maximum number of person images, identified by performance corresponding to process the person images during the time interval.