Human Counting via Bayesian Fusion of Detection Modalities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human detection techniques in images suffer from high false prediction rates and lack accuracy when multiple methods are combined, particularly in real-time applications like security and surveillance, where reliable human counting is crucial.
Innovation Solution
A system and method that utilizes a Bayesian fusion technique to integrate multiple human detection modalities, such as Histogram Oriented Gradient (HOG), Haar, and Background Subtraction, to calculate activity and accuracy probabilities, selectively choosing the most accurate modality for enhanced human counting in real-time images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple human detection modalities are combined to improve detection coverage, then the detection capability is enhanced, but the false prediction rate increases
Solution Approach 1:
The patent combines multiple human detection modalities (Haar, HOG, background subtraction) into a unified detection system. Each modality detects different aspects of human presence, and their results are merged through a fusion mechanism that aggregates detection outcomes across all modalities to improve overall detection capability.
Solution Approach 2:
The system calculates accuracy probability for each detection modality based on activity probability and detection results. This accuracy information feeds back into the fusion process, allowing the system to dynamically adjust the weight or trust given to each modality's detection results, thereby reducing false predictions by relying more on reliable modalities.
2Reliability
If multiple human detection modalities are used to reduce false predictions, then detection reliability improves, but system complexity increases
Solution Approach 1:
The detection system is segmented into independent modalities (Haar, HOG, background subtraction), each handling specific detection tasks. This segmentation allows each component to be optimized independently and simplifies the overall system architecture by dividing the complex detection problem into manageable sub-tasks that can be processed separately and then combined.
Solution Approach 2:
The patent introduces an intermediary fusion mechanism that mediates between multiple detection modalities and the final detection output. This intermediary layer calculates accuracy probabilities and integrates results from different modalities in a systematic way, reducing the direct complexity of coordinating multiple modalities while maintaining high detection reliability.
3Measurement precision
If accuracy probability calculation is performed for each modality, then detection accuracy is enhanced, but computational overhead increases
Solution Approach 1:
The system calculates accuracy probability selectively rather than for all modalities in all conditions. By using activity probability to determine when and which modalities require accuracy calculation, the system performs partial action only where necessary, reducing unnecessary computational overhead while maintaining high detection accuracy where it matters most.
Data Source
AI summary
The present invention discloses a method and a system for enhancing accuracy of human counting in at least one frame of a captured image in a real-time in a predefined area. The present invention detects human in one or more frames by using at least one human detection modality for obtaining the characteristic result of the captured image. The invention further calculates an activity probability associated with each human detection modality. The characteristic results and the activity probability are selectively integrated by using a fusion technique for enhancing the accuracy of the human count and for selecting the most accurate human detection modality. The human is then performed based on the selection of the most accurate human detection modality.