Coarse-to-Fine Hand Detection Using Deep Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CNN-based hand detection techniques are time-consuming and require significant computing power due to the processing of large amounts of irrelevant image data, which can be improved by using a coarse location network to initially locate hand segments and then a fine grid-based detection network for precise hand detection.
Innovation Solution
A two-stage approach involving a simple cascade CNN for coarse hand location and a grid-based detection network for fine hand detection, where the coarse location network reduces the input data to the fine grid-based network by focusing on areas predicted to contain hands, thereby reducing computational load and enhancing accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep CNNs are used for hand detection to achieve higher accuracy, then detection accuracy is improved, but computing power requirement increases and detection time is extended
Solution Approach 1:
The patent segments the hand detection process into two distinct stages: a coarse detection stage that quickly identifies potential hand regions, and a fine detection stage that precisely locates hands within those regions. This segmentation allows the system to process only relevant image portions in the detailed analysis phase, significantly reducing computational load and detection time while maintaining high accuracy through specialized processing at each stage
2Measurement precision
If deep CNNs are used for hand detection to achieve higher accuracy, then detection accuracy is improved, but computing power requirement increases
Solution Approach 1:
The patent extracts and processes only the relevant portions of image data identified by the coarse detection stage. By taking out the hand-containing regions from the full image for detailed CNN processing, the system minimizes the amount of data requiring intensive computational resources, thereby reducing energy consumption while preserving detection accuracy for the actual hand regions
3Reliability
If manual annotation is used to build training data for hand detection, then detection reliability is improved, but data preparation time and cost increase
Solution Approach 1:
The patent performs preliminary coarse detection to identify potential hand regions before detailed analysis. This preliminary action creates a focused set of regions that require minimal manual annotation, reducing the overall time and effort needed for training data preparation while maintaining reliable detection performance through the two-stage processing approach
Data Source
AI summary
Embodiments provide a process to identify one or more areas containing a hand or hands of one or more subjects in an image. The detection process can start with coarsely locating one or more segments in the image that contain portions of the hand(s) of the subject(s) in the image using a coarse CNN. The detection process can then combine these segments to obtain the one or more areas capturing the hand(s) of the subject(s) in the image. The combined area(s) can then be fed to a grid-based deep neural network finely detect area(s) in the image that contain only the hand(s) of the subject(s) captured.