Stereo Parallax Segmentation for Haze Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing stereo parallax calculating methods, such as SGM, face challenges in accurately determining whether an object with low image edge strength is haze, often leading to erroneous classifications and complex determination processes.
Innovation Solution
An object recognizing apparatus and method that acquire images from a stereo camera, calculate parallax, extract segments within a predetermined range, and determine if the variation in upper end positions and height widths of these segments exceeds specific thresholds to identify haze.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If stereo parallax calculating methods such as SGM are used to calculate parallax in areas with low image edge strength, then parallax can be calculated in walls or guardrails with low surface irregularities, but the parallax tends to vary greatly and objects are possibly erroneously determined to be haze
Solution Approach 1:
The patent applies local quality by analyzing the variation amount of parallax within local segments (continuous regions in the parallax image) rather than using global parallax values. By dividing the parallax image into segments and calculating the variation amount of parallax within each segment, the method can locally identify haze regions while maintaining accurate parallax calculation in non-haze areas with low image edge strength.
2Measurement precision
If haze determination is based on size change on time series, then haze can be identified, but the determination requires certain time and the process becomes highly complicated
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing the parallax image from stereo vision before haze determination is needed. The method also pre-defines segmentation criteria and variation amount thresholds. When haze determination is required, these pre-computed data and criteria are immediately used, eliminating the need for time-consuming real-time calculations and complex multi-step processes.
3Adaptability or versatility
If objects with low image edge strength are detected using stereo parallax methods, then more objects can be detected including those with low surface irregularities, but it is hard to accurately determine whether they are haze
Solution Approach 1:
The patent introduces an intermediary approach by using segment-based parallax variation analysis as a mediator between raw parallax data and haze determination. Instead of directly classifying objects based on parallax values alone, the method uses segments as intermediaries to analyze local parallax variations, providing a bridge that enables accurate haze identification while maintaining broad object detection capability.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach allows for accurate determination of haze in images, even for objects with low image edge strength, preventing erroneous classification and simplifying the determination process.
Implementation Method 1
calculating means configured to calculate parallax from the plurality of images
Data Source
AI summary
An object recognizing apparatus is provided with: an acquirer configured to acquire a plurality of images photographed by a stereo camera; a segment extractor configured to calculate parallax from the plurality of images and to extract a set in which the parallax is within a predetermined range, as a segment, at intervals of a certain width in an image lateral direction; a target extractor configured to extract segments coupled in the image lateral direction and an image depth direction, as a target; a calculator configured to calculate a variation amount of upper end positions and a variation amount of height widths of the segments that constitute the target; and a determinator configured to determine that the target is a haze if at least one of the variation amount of the upper end positions and the variation amount of the height widths is greater than a predetermined threshold value.