Optical Flow Confidence Map Using Decision Tree Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical flow estimation methods face challenges in accuracy and reliability, particularly in scenarios with occlusions, camera rotations, low visibility, and high motion, which can lead to incorrect flow estimates, and lack a real-time confidence map to assess the correctness of flow values.
Innovation Solution
A method for generating a confidence map by computing a set of features for each pixel, including image and flow gradients, and using a decision tree classifier to assign weights, allowing higher-level algorithms to assess and adjust for potentially incorrect optical flow estimates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional optical flow estimation methods are used, then computational speed is achieved, but accuracy deteriorates in challenging scenarios such as occlusions, camera rotations, low visibility, and high motion
Solution Approach 1:
The patent applies local quality by computing confidence values on a pixel-by-pixel basis rather than uniformly across the entire image. The confidence map is generated with different confidence levels for different regions, allowing the system to adaptively weight or discard flow estimates based on local image characteristics such as occlusions, boundaries, and motion patterns.
Solution Approach 2:
The patent segments the optical flow estimation process into multiple components: forward flow estimation, backward flow estimation, confidence map generation, and quality-based filtering. This segmentation allows each component to be optimized independently and combined to achieve high accuracy while maintaining computational efficiency through selective processing.
2Reliability
If confidence map computation is added to assess flow correctness, then reliability improves, but computational overhead increases
Solution Approach 1:
The patent performs preliminary action by computing the confidence map simultaneously with the optical flow estimation rather than as a separate post-processing step. The confidence values are derived from the same image gradients and flow computations already performed, avoiding redundant calculations and enabling real-time reliability assessment without significant additional overhead.
3Measurement precision
If higher level algorithms use confidence weights to discard incorrect flows, then accuracy improves, but processing complexity increases
Solution Approach 1:
The patent implements feedback by using the confidence map to guide higher-level algorithms in weighting or discarding flow estimates. The confidence values provide continuous feedback about the reliability of each pixel's flow measurement, allowing adaptive processing where high-confidence regions are processed with standard algorithms while low-confidence regions trigger alternative processing or are discarded.
Data Source
AI summary
A confidence map for optical flow gradients is constructed calculating a set of gradients for each pixel of an image, filtering said gradients and extracting confidence values from said gradients using a plurality of decision tree classifiers. A confidence map is then generated from said confidence values.