Multi-Camera Video Processing Using Epipolar Line Cropping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-camera video systems for human pose estimation are slow and require significant processing power, especially when dealing with moving objects and platforms, and they often rely on AI-based object detection systems that are not optimized for real-time processing.
Innovation Solution
The system employs an AI-based object detection engine and pose estimation engine to quickly identify and crop video data around moving objects using epipolar lines, reducing the computational load by processing only the necessary frames, and can be applied to both AI-based and non-AI systems, including those on moving platforms and non-video cameras.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI-based object detection systems are used by all cameras in a multi-camera system, then object identification accuracy is improved, but processing speed deteriorates and processing power requirements increase
Solution Approach 1:
The patent divides the processing task into segments: first perform fast epipolar line-based cropping to identify regions of interest, then apply AI-based object detection only to these cropped regions rather than processing all video data from all cameras with AI systems. This segmentation allows accurate object identification while significantly reducing processing time and computational power requirements.
Solution Approach 2:
The patent performs preliminary cropping of video data using epipolar lines before applying AI-based object detection. This preliminary action reduces the amount of data that subsequently requires AI processing, thereby improving overall processing speed while maintaining identification accuracy through the two-stage approach.
2Measurement precision
If AI-based object detection systems are used by all cameras, then object detection capability is improved, but processing power requirements increase
Solution Approach 1:
The system segments processing requirements by applying computationally lightweight epipolar line calculations to all camera feeds, then applying power-intensive AI-based object detection only to the small subset of cropped regions where objects are likely to be found. This dramatically reduces total processing power consumption while preserving detection capability.
Solution Approach 2:
Instead of applying full AI-based object detection to all camera feeds (excessive action), the system applies partial processing (epipolar line cropping) to all feeds and reserves AI processing only for the partial subset of regions that contain potential objects, thereby reducing processing power requirements while maintaining detection effectiveness.
3Reliability
If conventional processing methods are used, then processing accuracy is maintained, but processing time increases
Solution Approach 1:
The patent performs preliminary epipolar line-based cropping to pre-identify and isolate regions containing moving objects before subsequent processing steps. This preliminary action reduces the data volume for later accurate processing, thereby decreasing processing time while maintaining reliability through the preserved spatial relationships established by epipolar geometry.
Data Source
AI summary
A system and method for processing video data of an object in a movement space comprises uses a plurality of cameras each facing the movement space. An object detection engine running on a processor is in communication with each of the plurality of cameras that transmit video data of the movement space to the processor. The object detection engine determines the coordinates of a bounding box that closely surrounds the object in a frame of the video data from a master camera. The processor then determines, epipolar lines corresponding to points on the bounding box in the frame of the master camera in each of the frames of the video data from the non-master cameras. The processor crops the video data in the frame of the non-master cameras outside two of the epipolar lines to reduce and improve the video data processing.