Depth Image Noise Filtering via IR Intensity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional depth image filtering methods are limited in effectively removing noise due to their reliance on pixel unit comparisons and adjacent pixel filtering, which restricts the accuracy of noise removal in depth images acquired using a time of flight scheme.
Innovation Solution
An image filtering apparatus and method that calculates standard deviations of depth values and average infrared (IR) intensity values to generate a noise prediction model, allowing for enhanced noise removal by adjusting filter parameters such as noise enhancement folds, search ranges, and block similarities within the depth image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional pixel unit comparison and adjacent pixel filtering is used, then the filtering process is simple, but the noise removal accuracy is insufficient
Solution Approach 1:
The patent changes the filtering approach from simple spatial comparison to a model-based method that incorporates multiple parameters including IR intensity, standard deviation of depth values, and noise prediction models. This allows adaptive filtering strength adjustment based on local noise characteristics, improving noise removal accuracy while managing complexity through systematic parameter integration
Solution Approach 2:
The patent introduces an intermediary noise prediction model that mediates between the raw depth image and the filtered output. This model uses IR intensity and depth value statistics to predict noise levels, enabling more accurate noise removal by acting as an intermediate processing layer that guides the filtering operation
2Adaptability or versatility
If a fixed filtering approach is used, then the processing is fast, but it cannot adapt to varying noise levels in different image regions
Solution Approach 1:
The patent applies local quality by computing region-specific parameters including local standard deviation of depth values and local average IR intensity. This enables the filtering to adapt to varying noise levels in different image regions, as each region is processed according to its own statistical characteristics rather than a global fixed parameter
Solution Approach 2:
The patent introduces dynamics by making filter parameters adaptive rather than static. The noise prediction model dynamically adjusts filtering strength based on real-time calculation of standard deviation and IR intensity metrics, allowing the system to respond to varying noise conditions across different regions and frames
3Measurement precision
If more depth images and IR intensity images are used for calculation, then the noise prediction accuracy is improved, but the computational load increases
Solution Approach 1:
The patent applies partial action by selectively processing only the necessary statistical parameters (standard deviation and average IR intensity) from multiple images rather than analyzing all possible image features. This provides sufficient noise prediction accuracy while avoiding excessive computational burden from processing redundant information
Data Source
AI summary
Provided is an image filtering apparatus and method that may generate a noise prediction model associated with a depth image, based on an infrared ray (IR) intensity, and thereby predict noise included in a depth image using the generated noise prediction model.