Stereo Matching Parallax Selection for Backlit Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In environment recognition systems using stereo cameras, backlit environments disrupt the brightness balance between camera pairs, leading to increased SAD values and ineffective parallax calculation, making it difficult to detect objects, especially when edge processing loses information and amplifies noise.
Innovation Solution
The system conducts stereo matching on multiple pairs of images taken or processed differently, dividing distance images into sections to calculate representative parallaxes, which are then selected based on various criteria such as frequency, variance, and noise removal thresholds, ensuring accurate object detection even in challenging conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If stereo matching is performed using conventional SAD calculation on a single pair of images, then the process is simple and fast, but the object detection reliability deteriorates in backlit environments due to brightness imbalance
Solution Approach 1:
The patent divides the distance image into multiple vertical strip sections and calculates representative parallaxes for each section. This segmentation allows the system to process multiple image pairs with different processing methods (original images, edge-detected images, brightness-adjusted images) and select the most reliable parallax data from appropriate sections, thereby improving overall detection reliability without requiring complete processing of all image pairs.
Solution Approach 2:
The patent applies different image processing parameters and methods to create multiple pairs of images: edge detection parameters, brightness adjustment parameters, and different SAD calculation thresholds. By changing these parameters across multiple image pairs, the system can find at least one pair that provides reliable parallax calculation even in backlit conditions, improving object detection reliability.
2Measurement precision
If multiple pairs of images are processed with different image processing methods, then the representative parallax selection accuracy improves, but the processing time and computational load increase
Solution Approach 1:
By dividing the distance image into vertical strip sections and calculating representative parallaxes for each section, the system can process multiple image pairs in parallel and select the most accurate parallax data from appropriate sections. This segmentation reduces the computational burden compared to processing every pixel in every image pair, thereby reducing processing time while maintaining high accuracy.
Solution Approach 2:
The patent processes multiple pairs of images with different processing methods (edge detection, brightness adjustment, original images), which is excessive action beyond the minimum single-pair processing. However, by selecting only the necessary representative parallaxes from the processed sections, the system achieves high measurement precision while controlling processing time through selective rather than exhaustive use of all processed data.
3Measurement precision
If edge processing is applied to enhance object edges, then the object boundary detection improves, but information is lost and noise is amplified
Solution Approach 1:
The patent creates multiple pairs of images with different processing methods: one pair uses edge detection for boundary precision, another pair uses brightness adjustment for overall visibility, and a third pair uses original images for complete information. By making the system universal in handling multiple processing types, it can select from different sources depending on the situation, thereby maintaining object boundary detection precision while avoiding information loss by not relying solely on edge processing.
Solution Approach 2:
The patent combines multiple image processing approaches (edge detection, brightness adjustment, original images) into a composite processing system. Instead of using a single processing method that has drawbacks, the composite approach allows the system to leverage the strengths of each method while compensating for their weaknesses, achieving accurate boundary detection without significant information loss.
4Quantity of substance
If the SAD threshold is set low to accept more matches, then the parallax calculation coverage increases, but the measurement accuracy decreases due to including unreliable matches
Solution Approach 1:
The patent applies different SAD threshold criteria to different vertical strip sections. Instead of using a uniform threshold across the entire image, the system can set lower thresholds in sections where multiple image pairs provide consistent results and higher thresholds in sections where reliability is more critical. This local quality approach increases the overall number of matched pixel blocks while maintaining measurement precision through section-specific threshold optimization.
Solution Approach 2:
The system calculates representative parallaxes for multiple image pairs and uses the consistency and distribution of these parallaxes across different sections as feedback to determine appropriate threshold settings. By analyzing the frequency and variance of parallax values, the system can dynamically adjust threshold levels to maximize the number of valid matches while maintaining measurement accuracy, resolving the contradiction between quantity and precision.
Data Source
AI summary
An environment recognition system includes image taking means for taking a pair of images of an object in a surrounding environment with a pair of cameras and outputting the pair of images, stereo matching means for conducting stereo matching on a plurality of pairs of images that are taken by different image taking methods or that are formed by subjecting the pair of taken images to different image processing methods and forming distance images respectively for the pairs of images, selection means for dividing the distance images into a plurality of sections, calculating representative parallaxes respectively for the sections, and selecting any of the representative parallaxes of the corresponding section as a representative parallax of the section, and detection means for detecting the object in the image on the basis of the representative parallaxes of the sections.