Feature Point Detection Model with Parallel Processing Heads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object joint detection techniques face challenges in improving accuracy and reducing calculation load, as they require dividing the image into regions for processing, leading to a heavy computational burden.
Innovation Solution
An object feature point detection device that includes a detection model, a learning part, and a parameter optimization part, which outputs estimated positions for feature points without dividing the image, using optimized parameters to calculate total sum errors and associate them with correct answer positions for efficient learning and reduced calculation load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the image is divided into regions for object joint detection processing, then the detection can be performed for specific regions, but the calculation load becomes heavy
Solution Approach 1:
The detection model is segmented into multiple processing heads (e.g., keypoint detection head, object detection head) that operate in parallel. Each head processes specific aspects of the input image independently, allowing the system to maintain high detection precision while distributing the calculation load across multiple specialized modules rather than processing the entire image uniformly
Solution Approach 2:
The patent transforms the detection problem from a 2D image space search into a combined 2D spatial + 1D feature space approach. By using pre-trained backbone networks to extract features and then applying detection heads in the feature space, the system reduces the effective search space and calculation requirements while maintaining or improving detection accuracy
2Measurement precision
If the image is finely divided to improve position estimation accuracy, then the accuracy improves, but the calculation load becomes heavier
Solution Approach 1:
The backbone network performs preliminary feature extraction and processing before the detection heads operate. This preliminary action of extracting meaningful features from the input image allows subsequent detection steps to work with compressed, informative representations rather than raw pixel data, improving both accuracy and efficiency
Solution Approach 2:
The system changes parameters such as feature map resolution, number of detection layers, and processing granularity dynamically. By adjusting these parameters based on the specific detection task and available computational resources, the system can optimize the balance between position estimation accuracy and processing efficiency for different scenarios
3Reliability
If the entire region of the image is searched for object joint detection, then comprehensive detection is achieved, but the calculation load increases
Solution Approach 1:
The detection task is segmented into multiple independent detection heads (keypoint detection, object detection, relationship detection) that process different aspects of the image simultaneously. This segmentation allows comprehensive detection across the entire image region while distributing computational work across parallel processing streams, maintaining detection completeness without concentrating all calculation load in a single processing pipeline
Data Source
AI summary
Provided is an object feature point detection device including a detection model, a learning part, and a parameter optimization part. The detection model outputs estimated data including an estimated position of each feature point included in each of a plurality of objects in an input image. The learning part executes the detection model and machine learning of the detection model. The parameter optimization part optimizes a parameter for estimating a position of each feature point of an optional object among objects of a plurality of pieces of estimated data output from the detection model to which a training image obtained by imaging a plurality of objects is input. The detection model outputs estimated data including an estimated position of each feature point included in each of a plurality of objects in a newly input image by using the parameter optimized by the parameter optimization part.