3D Object Detection via Binarized Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing three-dimensional object detection methods require significant computational effort and resources, especially when detecting small objects, due to the need for large correlation windows and extensive point comparisons, which can lead to increased costs and difficulty in real-time processing.
Innovation Solution
A method that applies a binarization scheme to correlation and search windows using randomly selected point pairs to form resultant vectors, allowing for smaller correlation windows and reduced computational demands, enabling reliable detection of small objects with lower hardware and calculation costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If large correlation windows are used to provide good correlation between camera image and reference image, then detection reliability is improved, but object resolution deteriorates and calculation demand increases
Solution Approach 1:
The patent applies segmentation by dividing the correlation window into multiple sub-windows and processing them independently. This allows the system to use smaller effective correlation regions while maintaining overall detection reliability, thereby preserving object resolution without sacrificing correlation accuracy.
Solution Approach 2:
The patent uses partial action by selectively processing only certain regions or features within the correlation window rather than the entire window. This reduces the effective computation required while maintaining detection reliability for the most important features, thus resolving the contradiction between reliability and computational demand.
2Reliability
If large correlation windows are used to provide good correlation, then detection reliability is improved, but calculation demand increases significantly
Solution Approach 1:
By segmenting the correlation window into sub-windows, the patent reduces the computational complexity from O(n²) to O(k*n²/m) where k is the number of sub-windows and m is the reduction factor. This enables real-time processing while maintaining correlation reliability through the combined results of multiple sub-window correlations.
Solution Approach 2:
The patent applies partial action by focusing computational resources on the most informative regions of the correlation window. This selective processing maintains detection reliability while significantly reducing the overall calculation demand, enabling real-time processing capabilities.
3Measurement precision
If many point comparisons are performed in the census process, then correlation accuracy is improved, but calculation effort and cost increase
Solution Approach 1:
The patent segments the set of point comparisons into multiple subsets corresponding to different sub-windows. This reduces the number of comparisons required in each subset while maintaining overall correlation accuracy through the aggregation of results from all sub-windows, thereby reducing calculation effort and cost.
Solution Approach 2:
The patent uses partial action by performing point comparisons only on selected subsets of points rather than all possible point pairs. This selective approach maintains correlation accuracy for the most discriminative features while significantly reducing the total number of comparisons and associated computational cost.
Data Source
AI summary
The present invention relates to a method for the three-dimensional detection of objects, in which an image is detected and is compared to a known reference image and mutually corresponding objects in the images are identified by means of a correlation process. A binarization scheme is used in the correlation process and compares randomly selected point pairs to one another. The point pairs are fixed by means of a plurality of iterations.