Modified SGM Optical Flow Mapping for Accurate Motion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining optical flow are inefficient and lack accuracy in processing image data for applications such as motion estimation, object detection, and robot navigation.
Innovation Solution
A modified Semi-Global Matching (SGM) method is employed to generate optical flow maps by processing reference and input images, utilizing denoising, edge detection, object detection, and decision logic to determine disparity maps, which enhance the accuracy and efficiency of optical flow computation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional optical flow methods are used, then computational simplicity is maintained, but accuracy and precision of motion estimation deteriorate
Solution Approach 1:
The patent segments the optical flow computation into distinct processing stages: edge detection identifies boundary regions, object detection locates significant features, and disparity map generation computes depth information. Each stage processes specific image regions independently, improving accuracy while managing computational load through divided responsibility.
Solution Approach 2:
The patent performs preliminary processing steps before main optical flow computation: denoising cleans the input images, edge detection pre-identifies boundary regions, and object detection pre-locates significant features. These preliminary actions prepare the data structure and reduce the search space for subsequent disparity computation, improving both accuracy and efficiency.
2Measurement precision
If conventional optical flow methods are used, then processing speed is maintained, but precision of disparity determination deteriorates
Solution Approach 1:
The patent applies different processing qualities to different image regions: edge regions receive enhanced processing with dedicated edge detection and boundary handling, object regions receive focused object detection and tracking, and background regions receive standard processing. This local quality differentiation improves disparity precision in critical regions without uniformly increasing processing load across the entire image.
Solution Approach 2:
The patent introduces intermediate data structures and processing layers: edge maps serve as intermediaries between raw images and disparity computation, object masks act as intermediaries for region-of-interest processing, and cost aggregation functions as intermediaries between local matching and final optical flow. These intermediaries refine the information progressively, improving precision at each stage.
3Measurement precision
If image processing steps are added for denoising and edge detection, then optical flow accuracy is improved, but computational time increases
Solution Approach 1:
The patent merges multiple processing functions into integrated pipelines: denoising is combined with edge detection in a unified preprocessing stage, object detection is merged with disparity computation in region-of-interest processing, and cost aggregation is combined with optical flow generation. These merges reduce redundant operations and improve computational efficiency while maintaining accuracy enhancements.
Solution Approach 2:
The patent applies enhanced processing selectively rather than uniformly: full denoising and edge detection are applied only to regions where they most impact accuracy, object detection is performed at varying levels of detail across different image regions, and disparity computation uses adaptive precision based on local image characteristics. This partial application of processing steps improves accuracy where needed while minimizing unnecessary computational overhead.
Data Source
AI summary
Apparatuses, systems, and techniques to determine optical flow. In at least one embodiment, a set of disparity values is used to determine optical flow between input and reference images. For each of a plurality of image regions of the input image, the set of disparity values may include disparity values for a plurality of directions intersecting the image region.