Polygonal Sub-Image Cropping for Low Complexity Motion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing humanoid detection systems face challenges in achieving high detection accuracy in real-time due to increased computational complexity and hardware/software costs, making it difficult to implement in systems with low computing capability.
Innovation Solution
A motion detection method and system that acquires a raw image, detects motion objects using a motion detector, crops the image into a polygonal sub-image, and processes it with a neural network to determine matching detection categories, reducing computational complexity by analyzing only the sub-image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high detection accuracy and instant detection results are required, then computational complexity and image processing algorithm complexity must be increased, but this leads to increased hardware design costs and software testing costs
Solution Approach 1:
The patent divides the image processing task into two stages: first, a motion detector identifies regions with motion objects; second, only these identified regions are cropped and processed by the neural network. This segmentation of the processing task reduces the computational complexity from analyzing the entire high-resolution image to analyzing only relevant sub-regions, thereby resolving the contradiction between detection accuracy and computational complexity.
Solution Approach 2:
The patent extracts only the necessary sub-images containing motion objects from the full raw image using a motion detector and cropping mechanism. By taking out and processing only these extracted sub-regions rather than the complete image, the system achieves high detection accuracy with reduced computational complexity and lower hardware requirements.
2Measurement precision
If high detection accuracy is achieved through improved image processing algorithms for floating-point numbers, then detection performance improves, but hardware design costs and software testing costs increase
Solution Approach 1:
The patent extracts only the necessary sub-images containing motion objects from the full raw image using a motion detector and cropping mechanism. By taking out and processing only these extracted sub-regions rather than the complete image, the system achieves high detection accuracy with reduced computational complexity and lower hardware design costs.
Solution Approach 2:
The patent employs a motion detector as a preliminary filtering stage that identifies regions of interest, allowing the expensive neural network processor to operate only on small sub-images rather than full-resolution images. This disposable pre-processing approach reduces the computational burden on the main processing system, lowering hardware design costs while maintaining detection accuracy.
3Measurement precision
If the entire raw image is processed by the neural network, then detection accuracy may be improved, but processing time increases and real-time detection becomes difficult
Solution Approach 1:
The patent segments the image processing workflow into motion detection and classification stages, with only motion-containing sub-regions being processed by the neural network. This time-efficient segmentation reduces processing time from analyzing the entire raw image to analyzing only relevant sub-images, enabling real-time detection while maintaining accuracy.
Solution Approach 2:
The patent performs preliminary motion detection and image cropping before neural network classification. This preliminary action of identifying and extracting only regions containing motion objects reduces the amount of data requiring expensive neural network processing, thereby reducing processing time and enabling real-time detection performance.
Data Source
AI summary
A motion detection method includes acquiring a raw image, detecting a motion object image according to the raw image by using a motion detector, cropping the raw image to generate a sub-image according to the motion object image, and inputting the sub-image to a processor for determining if a motion object of the sub-image matches with a detection category. The processor includes a neural network. The shape of the sub-image is a polygonal shape.