Mixed Distribution Model for Moving Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for detecting moving objects in moving images face challenges in accurately setting background values, especially in the initial stages and during updates, leading to unstable detection accuracy due to insufficient data and potential outliers.
Innovation Solution
An image processing apparatus that generates and updates a mixed distribution model for each pixel by incorporating surrounding pixel values, allowing for accurate background value determination and moving object detection, even in noisy environments, by using pixel values from local regions across frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If a mixed distribution model is generated from only the first pixel value without using data on a large number of actual pixel values, then the model can be generated in the initial stage, but the detection accuracy is unstable and erroneous detection occurs
Solution Approach 1:
The patent merges spatial information from surrounding pixels with temporal information from the current pixel value across frames. When generating a distribution model for a pixel, it combines data from multiple surrounding pixels in the same frame and the current pixel value from previous frames, creating a more robust model that works well in both initial and update stages
Solution Approach 2:
The patent performs preliminary action by pre-generating distribution models using surrounding pixel values before actual detection begins. This allows the system to have ready-to-use background models from the first frame, eliminating the need to wait for accumulation of temporal data while maintaining detection accuracy
2Measurement precision
If pixel values from surrounding pixels in the same frame are used to generate a distribution model, then the model can be generated in the initial stage, but it may include noise from other objects
Solution Approach 1:
The patent applies local quality by selectively using surrounding pixel values only within a defined local region around the target pixel. This localized approach ensures that only pixels from the same background region are incorporated, excluding pixels from other objects or distant regions that would introduce noise
Solution Approach 2:
The system uses feedback mechanisms to evaluate whether surrounding pixel values should be incorporated into the distribution model. By analyzing the consistency and reliability of surrounding pixel data, the system dynamically adjusts which pixels to include, reducing noise while maintaining accuracy
Data Source
AI summary
The image processing apparatus for detecting a moving object in a moving image includes a background generation unit configured to generate a background image of the moving image while updating the background image over time. The background generation unit includes a model derivation unit configured to derive a mixed distribution model having one or more distribution models for each pixel of interest, and a background value derivation unit configured to derive one or more background pixel values respectively corresponding to the one or more distribution models. The model derivation unit is configured to generate a new distribution model from pixel values of a plurality of pixels within a local region containing the pixel of interest in a first frame, and update the existing distribution model using a pixel value of the pixel of interest in a second frame that is different from the first frame.