Image Processing Apparatus Object Likelihood Distribution Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image tracking methods face challenges in accurately tracking specific objects due to insufficient feature amounts when using small template images and incorrect inclusion of background elements when using larger templates, especially in scenarios where distance information is difficult to detect, leading to reduced tracking accuracy.
Innovation Solution
An image processing apparatus that acquires distance information from first image data, generates a likelihood distribution based on object position and distance information, and calculates feature amounts for object detection in second image data, using parallax images to refine tracking accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If a small image region is used as template image, then tracking speed is improved, but feature amount becomes insufficient for accurate tracking
Solution Approach 1:
The patent segments the image into multiple depth layers based on distance information from the camera. Instead of using a single template image, it creates multiple template images corresponding to different depth ranges. This allows the system to maintain small template sizes for fast processing while incorporating sufficient feature information by selecting appropriate depth layers, thus resolving the contradiction between tracking speed and accuracy.
Solution Approach 2:
The patent introduces a depth dimension to the traditional 2D template matching approach. By organizing templates across multiple depth layers rather than using a single 2D template, the system gains additional dimensional information that enhances feature recognition capability without increasing the spatial size of individual templates, thereby maintaining fast tracking speed while improving accuracy.
2Measurement precision
If a large image region is used as template image, then feature amount is sufficient for accurate tracking, but background elements are incorrectly included
Solution Approach 1:
The patent segments the image region into multiple depth layers based on distance information. By creating separate template images for different depth ranges, the system can select templates that correspond to the actual object depth, excluding background elements at different depths. This segmentation approach allows sufficient feature information to be captured while preventing background contamination, resolving the contradiction between tracking accuracy and background exclusion.
Solution Approach 2:
The patent applies different depth range assignments to different regions of the image based on their actual content and distance characteristics. By assigning appropriate depth layers to specific spatial regions, the system ensures that each template image contains only relevant features for objects at that depth, preventing background elements from being incorrectly included while maintaining sufficient feature information for accurate tracking.
3Measurement precision
If distance information is used to prevent background inclusion, then tracking accuracy is improved, but processing load increases
Solution Approach 1:
The patent performs preliminary action by pre-calculating and storing depth information for the entire image before the tracking process begins. The image is divided into multiple depth layers in advance, and this pre-organized depth information is readily available during tracking without requiring real-time complex calculations. This preliminary processing reduces the computational burden during actual tracking operations while maintaining high accuracy through distance-based background exclusion.
4Measurement precision
If distance information is calculated for the entire image, then tracking accuracy is improved, but processing load increases
Solution Approach 1:
The patent segments the image into multiple depth layers and processes each layer independently. Instead of calculating distance information for the entire image as a single large task, the system divides the computation into smaller, manageable segments corresponding to different depth ranges. This segmentation allows accurate distance-based tracking while improving processing efficiency through parallelization and reduced computational complexity for each individual layer.
Data Source
AI summary
One aspect of the present invention provides an image processing apparatus capable of tracking an image region to be tracked with high accuracy.A distance distribution generation unit generates distance information indicating a distance of each of regions in first image data. An object likelihood distribution generation unit generates a likelihood distribution of an object in the first image data based on information indicating a position specified as the object in the first image data, the distance information, and reliability of the distance information. A feature amount extraction unit calculates a feature amount to be used to detect the object from second image data based on the likelihood distribution.