Statistical False Detection Removal Algorithm for Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI-based object detection technologies face challenges in efficiently correcting false detections during model learning, requiring significant resources and time for manual indexing and re-learning, especially when trained with limited data.
Innovation Solution
A method for false detection removal in image processing devices, involving the use of two filtering models: a feature-based model that extracts and models the distribution of an object's feature vector, and a color-based model that analyzes primary colors in the CIE-LAB color space, to identify and correct erroneously detected objects, thereby improving detection performance with minimal data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual indexing and re-learning are used to correct false detections, then detection reliability is improved, but time consumption and resource usage increase significantly
Solution Approach 1:
The system automatically identifies and corrects false detections through self-service mechanisms. The false detection filtering model autonomously processes detected objects, comparing their features against learned distributions to identify and remove false positives without requiring manual intervention, thereby maintaining high reliability while minimizing time loss
Solution Approach 2:
The system performs preliminary action by pre-training false detection filtering models with expected object feature distributions before actual detection occurs. This advance preparation enables rapid automatic filtering of false detections during operation, eliminating the need for time-consuming manual correction while maintaining detection reliability
2Measurement precision
If extensive re-training is performed to correct false detections, then detection accuracy is improved, but computational resources and training time are consumed
Solution Approach 1:
The system extracts only the essential feature distributions needed for false detection identification rather than performing extensive re-training. By extracting and modeling key features (such as color histograms, texture patterns, or shape characteristics) into compact distributions, the system achieves accurate false detection filtering with minimal computational resource consumption
Solution Approach 2:
The system changes parameters by representing objects through their feature distributions (statistical parameters) rather than requiring full re-training of detection models. This parameter-based approach allows the system to adapt to false detections by adjusting distribution parameters efficiently, improving detection accuracy without consuming extensive computational resources
3Productivity
If simple techniques with small data amounts are used for training, then training efficiency is improved, but detection performance may be compromised
Solution Approach 1:
The system applies parameter changes by modeling feature distributions with statistical parameters (mean, variance, covariance) that can be accurately estimated from small data sets. This approach enables effective false detection filtering even when training data is limited, maintaining detection performance while achieving high training efficiency
Solution Approach 2:
The system uses composite materials by combining multiple feature types (color, texture, shape) into a composite feature distribution model. This composite approach leverages complementary information from different feature modalities, enabling robust false detection identification with small amounts of training data while maintaining high detection performance
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A false detection removal method of an image processing device are disclosed. In the present disclosure, after an object of interest is detected, removing feature-based false detection of the object of interest, removing color-based false detection of the object of interest are perfomed. According to the removing function, a final obkect of interest is obtained, detection performance may be improved even when an object of interest detector is trained with a simple technique using a small amount of data. In the present disclosure, one or more of a surveillance camera, an autonomous vehicle, a user terminal, and a server may be linked to an artificial intelligence module, a robot, an augmented reality (AR) device, a virtual reality (VT) device, and a 5G service.