Mobile Body Part Detection via Entropy Analysis and Dynamic Frame Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for body part detection and tracking on mobile devices are resource-intensive, leading to high CPU usage and false positives, and fail to efficiently adapt to changing conditions such as movement.
Innovation Solution
A system that uses reactive detection scheduling and entropy analysis to eliminate false positives by focusing processing on areas of high entropy, adjusting frame processing rates based on device and subject movement, and employing accelerometer and gyroscope data to optimize processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional body part detection methods are used, then detection capability is achieved, but CPU usage increases and false positives occur
Solution Approach 1:
The patent divides the detection process into multiple stages: initial candidate identification using simplified criteria, followed by progressive filtering through entropy analysis and template matching. This segmentation allows the system to process all frames at low cost while applying intensive processing only to promising candidates, resolving the contradiction between comprehensive detection and CPU conservation.
Solution Approach 2:
The patent applies different processing qualities to different regions and candidates: low-cost processing for all frames, medium-cost entropy analysis for potential candidates, and high-cost template matching only for high-probability candidates. This local quality approach ensures detection reliability while minimizing overall CPU usage by concentrating resources where they are most needed.
2Reliability
If comprehensive frame processing is performed, then detection reliability is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary entropy analysis on all frames before applying computationally intensive template matching. This preliminary action identifies and eliminates false positive candidates early in the pipeline, ensuring that time-consuming template matching is applied only to genuine candidates, thus improving false positive reduction without excessive processing time.
Solution Approach 2:
The patent maintains continuous monitoring of candidate regions across frames using tracking algorithms. Once a candidate is identified, the system continuously refines its position and probability without re-processing the entire frame, allowing detection reliability to improve over time while minimizing additional processing time through efficient temporal reuse of information.
3Adaptability or versatility
If fixed processing rate is used, then implementation simplicity is maintained, but adaptability to movement conditions deteriorates
Solution Approach 1:
The patent implements dynamic frame processing rates based on detected movement conditions. When motion is detected between frames, the system increases processing frequency to maintain tracking accuracy. When motion is minimal, processing rate decreases to conserve resources. This dynamic adaptation improves versatility across different movement conditions while keeping control complexity manageable through simple motion-detection-triggered adjustments.
Data Source
AI summary
A method is provided for detecting a body part in a video stream from a mobile device. A video stream of a human subject is received from a camera connected to the mobile device. The video stream has frames. A first frame of the video stream is identified for processing. This first frame is then partitioned into observation windows, each observation window having pixels. In each observation window, non-skin-toned pixels are eliminated; and the remaining pixels are compared to determine a degree of entropy of the pixels in the observation window. In any observation window having a degree of entropy above a predetermined threshold, a bounded area is made around the region of high entropy pixels. The consistency of the entropy is analyzed in the bounded area. If the bounded area has inconsistently high entropy, a body part is determined to be detected at that bounded area.