Object Detection Using Overlapping Sub-Regions and Cumulative Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection techniques, such as those using Haar-like features, are computationally demanding and challenging to implement in real-time on devices with limited processing resources, particularly in compact devices like digital cameras and mobile phones, where efficient face detection is necessary for auto-focus and exposure control.
Innovation Solution
A method and apparatus for object detection that performs tests on overlapping sub-regions of a search window, generates a cumulative score, and selectively continues testing based on threshold comparisons, utilizing a resize unit, image integration unit, and processing block that operate in parallel to optimize resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a cascade of tests is used for object detection, then detection accuracy is improved, but processing resource consumption increases
Solution Approach 1:
The patent divides the search window into multiple overlapping sub-regions and processes them in parallel. This segmentation allows the detection system to distribute the computational load across multiple independent test sequences, reducing the sequential processing burden while maintaining comprehensive coverage of the search space.
Solution Approach 2:
The patent performs tests on multiple overlapping sub-regions beyond what would be required for complete non-overlapping coverage. This excessive action ensures that potential objects near boundaries are not missed, while the parallel processing architecture mitigates the additional computational cost by efficiently utilizing available processing resources.
2Speed
If object detection is performed in real-time on compact devices, then responsiveness is improved, but processing resource constraints are worsened
Solution Approach 1:
By segmenting the search window into overlapping sub-regions that can be processed independently and in parallel, the patent reduces the computational complexity of processing the entire search window sequentially. This enables real-time detection on compact devices with limited processing resources.
Solution Approach 2:
The patent introduces parallel processing as an additional dimension of computation, transitioning from sequential single-threaded processing to multi-threaded or multi-core parallel execution. This dimensional change in processing architecture enables real-time performance despite hardware constraints.
3Productivity
If the number of tests is reduced for efficiency, then processing speed is improved, but detection accuracy deteriorates
Solution Approach 1:
The patent segments the detection task into multiple parallel test sequences operating on overlapping sub-regions. Each sequence performs a reduced number of tests efficiently, while the combination of results from multiple sequences maintains or improves overall detection accuracy through redundant coverage.
Solution Approach 2:
Different overlapping sub-regions may contain different numbers of objects or different types of objects. The patent allows each local region to be processed with appropriate testing depth, performing more thorough tests where objects are likely present and fewer tests in regions where objects are unlikely, thereby optimizing the balance between efficiency and accuracy.
Data Source
AI summary
The invention concerns a method of performing, by an image processing device, object detection in an image comprising: performing one or more tests of a test sequence for detection of a first object on pixels values of a plurality of at least partially overlapping sub-regions (310, 312, 314) of a first search window (108); generating a cumulative score based on results of said one or more tests on said plurality of sub-regions; comparing said cumulative score with a threshold value; and based on said comparison, selectively performing one or more of said tests of said test sequence on at least one further sub-region of said first search window, said at least one further sub-region at least partially overlapping each of said plurality of sub-regions.