Video Image Segmentation Using Affine Keypoint Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation methods in video processing consume a large quantity of computing resources due to the need for extensive data processing and calculations, particularly when using convolutional neural networks with optical flow images.
Innovation Solution
The method employs affine transformation based on prior location information of target object key points in previous frames to quickly determine the target object in current frames, reducing computational load by focusing on a region of interest and performing key point detection only on this region.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If convolutional neural network with optical flow images is used for segmentation, then segmentation accuracy is improved, but computing resource consumption increases
Solution Approach 1:
The patent divides the image processing task into two stages: first performing key point detection to identify target locations, then performing segmentation only in the vicinity of these key points. This segmented approach to processing different parts of the image at different detail levels reduces overall computational load while maintaining segmentation accuracy through focused analysis of critical regions.
Solution Approach 2:
The patent performs key point detection as a preliminary step before segmentation. By first identifying the locations of target objects through key point detection, the system can then focus segmentation resources only on relevant regions, avoiding the need to process the entire image with full segmentation complexity, thus reducing computing resource consumption.
2Measurement precision
If full image processing is performed for segmentation, then segmentation completeness is improved, but processing time increases
Solution Approach 1:
The patent applies different processing qualities to different regions: key point detection is performed on the full image to ensure completeness, while segmentation is performed with high detail only in local regions around detected key points. This local quality approach maintains segmentation completeness by covering all potential targets while reducing processing time through localized detailed analysis.
Solution Approach 2:
The patent segments the processing domain into key point detection phase (full image) and segmentation phase (local regions). This spatial segmentation of processing tasks ensures that no potential target is missed while avoiding unnecessary computation in regions without targets, thus balancing completeness and processing time.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The present application relates to an image segmentation method and apparatus, a computer device, and a storage medium. The method comprises: sequentially selecting a current image frame in a video according to a timing sequence; determining a reference image frame from image frames of which the timing sequences in the video are located in front of the current image frame; obtaining first position information of a target object key point in the reference image frame; carrying out affine transformation on the current image frame by referring to an affine transformation relation between the first position information and a target object key point template to obtain a target object diagram of the current image frame; performing key point detection on the target object diagram to obtain second position information of the target object key point; segmenting a target object from the target object diagram to obtain segmentation information of the target object; and segmenting the target object from the current image frame according to the segmentation information and the second position information. According to the solution of the present application, the calculated amount is reduced.