Object Detection Apparatus Using Machine-Learned Candidate Areas
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems face challenges in efficiently and accurately detecting objects in images from monitoring cameras, particularly due to the need for manual specification of candidate areas, which becomes impractical with increasing variations of objects and states, and fails to adapt to changes in the shooting environment over time.
Innovation Solution
An object detection apparatus that uses machine-learning to automatically calculate optimal candidate areas by analyzing past image data, reducing manual effort and adapting to environmental changes by regularly updating candidate areas based on new data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual specification of candidate areas is performed for each object and state, then detection accuracy is improved, but the complexity and time required increases significantly
Solution Approach 1:
The system automatically generates candidate areas and selects appropriate discriminators without manual intervention. The processor autonomously performs machine learning to create candidate area information and matches discriminators based on detection targets and states, eliminating the need for manual specification while maintaining detection accuracy.
Solution Approach 2:
The system dynamically adjusts candidate area parameters based on detection targets and their states. By changing the parameters of candidate areas according to the specific object and state being detected, the system achieves high detection accuracy without requiring manual configuration for each scenario.
2Measurement precision
If manual specification of candidate areas is performed for each object and state, then detection accuracy is improved, but the processing time increases
Solution Approach 1:
The system performs preliminary machine learning to generate candidate area information before actual detection occurs. By pre-processing and creating candidate area data in advance, the system reduces the time required during actual detection operations while maintaining high accuracy through pre-established candidate areas.
Solution Approach 2:
The automated system generates and updates candidate areas without manual intervention, significantly reducing the time required compared to manual specification. The processor autonomously performs machine learning and generates candidate area information, eliminating time-consuming manual operations.
3Measurement precision
If candidate areas are manually specified, then initial detection accuracy is improved, but adaptability to environmental changes deteriorates
Solution Approach 1:
The system dynamically updates candidate area information through continuous machine learning operations. Candidate areas are not static but are regularly updated based on new detection data and environmental changes, allowing the system to adapt to varying conditions while maintaining detection accuracy.
Solution Approach 2:
The system uses feedback from detection results to continuously improve candidate area information. By analyzing detection outcomes and using this feedback in subsequent machine learning operations, the system adapts to environmental changes and maintains high detection accuracy over time.
4Reliability
If sliding window method is applied to entire image, then comprehensive detection is achieved, but processing time increases
Solution Approach 1:
The system extracts and focuses only on relevant candidate areas within the image rather than processing the entire image. By using machine learning to identify and extract candidate areas where detection targets are likely to be found, the system reduces processing time while maintaining comprehensive detection through targeted analysis of specific regions.
Solution Approach 2:
Instead of applying the sliding window method to the entire image, the system applies it only to selected candidate areas. This partial action approach reduces processing time significantly while maintaining detection completeness by focusing computational resources on areas where targets are most likely to be found.
Data Source
AI summary
An object detection apparatus is provided with a discriminator applier and a candidate area calculator. The discriminator applier applies a discriminator which detects an object to images acquired in past and calculates object detection information including at least location information of the object detected by the discriminator, in a learning phase. The candidate area calculator performs a machine-learning by use of the object detection information and calculates object candidate area information including at least information specifying a candidate area in which the object may appear in an image.