Multi-Camera Human Action Detection via Predicted Motion Directions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting human actions in images captured by multiple cameras face challenges in processing load due to hidden or blurred body parts, leading to inaccuracies in posture estimation and action detection.
Innovation Solution
An image processing apparatus predicts actions of a person using multiple cameras, selects an optimal camera based on predicted actions, and uses machine learning to enhance posture estimation and action detection by prioritizing images with favorable orientations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If images are captured using a plurality of cameras to reduce the influence of hidden or blurred body parts, then the accuracy of posture estimation and action detection is improved, but the processing load increases
Solution Approach 1:
The system performs preliminary action prediction using a lightweight model before full posture estimation. By predicting actions in advance based on chronological changes in posture estimation results, the system can selectively process only relevant camera images, reducing the overall processing load while maintaining accuracy.
Solution Approach 2:
Instead of processing all camera images equally, the system applies partial processing by focusing computational resources on images that are most relevant for detecting predicted actions. This selective processing approach reduces the processing load while maintaining measurement precision.
2Reliability
If all camera images are processed for action detection, then comprehensive action detection is achieved, but the processing time increases
Solution Approach 1:
The system performs preliminary action prediction before full action detection. By predicting actions based on chronological posture changes and then selecting only relevant camera images for detailed analysis, the system reduces processing time while maintaining comprehensive action detection reliability.
Solution Approach 2:
The processing is segmented into two stages: first predicting actions using a lightweight model, then selectively processing relevant images with full detection algorithms. This segmentation reduces overall processing time while maintaining detection comprehensiveness.
Data Source
AI summary
An image processing apparatus according to one aspect of the present invention predicts the next action of a person who has been detected from an image captured by an image capturing apparatus, and selects an optimal image capturing apparatus for detecting the action of the person from among the plurality of image capturing apparatuses based on the predicted action.