Head Position Estimation Using Pre-Extracted Feature Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating the position of a head in images, such as those used in Human Computer Interaction, are inefficient due to time-consuming model generation and require high computational resources, making them unsuitable for applications like surveillance.
Innovation Solution
A computer-implemented method and apparatus that crops an image object representing a head, abstracts features using a predefined statistical model, calculates pixel probabilities, and determines the head position by computing the maximum log likelihood sum, reducing computational effort and enabling efficient head pose estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D model based tracking is used to estimate head position, then measurement precision is improved, but loss of time increases due to time-consuming model generation
Solution Approach 1:
The patent pre-extracts and stores facial feature point coordinates from a large database of images with known head poses during offline processing. This preliminary action eliminates the need for time-consuming 3D model generation during real-time application, as the system only needs to match stored feature points against new images using efficient algorithms like ICP or SIFT.
2Measurement precision
If machine learning techniques with high scale equipment are used to determine head position, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent segments the head detection task into independent components: extracting facial feature points (eyes, nose, mouth, ears) from images and matching them against a pre-stored database of feature point coordinates corresponding to known head poses. This segmentation allows the use of simple image processing algorithms rather than complex machine learning models, reducing device complexity while maintaining precision.
Solution Approach 2:
The patent creates a copy of facial feature point coordinates from a database of images with known head poses. Instead of using complex machine learning models to infer head position, the system copies and matches simplified feature point representations, significantly reducing computational requirements and device complexity.
3Measurement precision
If conventional head pose estimation methods are used, then measurement precision is improved, but productivity decreases due to high computational requirements
Solution Approach 1:
The patent uses simple, lightweight feature point matching algorithms instead of computationally intensive machine learning models. By matching discrete feature point coordinates using efficient algorithms like ICP or SIFT, the system achieves fast processing speeds suitable for real-time applications while maintaining acceptable precision for head pose estimation.
Data Source
AI summary
A computer implemented method and an apparatus for estimating head position of an individual are described. The method includes cropping an image object in an image. The image object may include a visual representation of a head of at least one individual. The method further includes abstracting one or more features from the image object. Subsequently,position of the head of the at least one individual in the image object is determined using the one or more features and a predefined statistical model. The predefined statistical model may include a value of each of the pixels of one or more stored images representing variety of head poses and ranges of head poses of the at least one individual.