Robot Vision Feature Extraction for Dynamic-Object SLAM
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Visual SLAM for robots is less robust with a higher proportion of dynamic objects, leading to positioning jitter and reduced accuracy due to the interference of dynamic objects, resulting in low stability and positioning information deviation.
Innovation Solution
A method involving the collection of image data from visual sensors and angular velocity data from inertial sensors to calculate relative poses between image frames, extract feature points, and project their positions, using algorithms like optical flow to filter out dynamic objects and maintain a stable number of feature points for accurate positioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If visual SLAM is used for robot positioning and navigation, then the cost is reduced and weight is decreased compared to laser-based SLAM, but the stability and positioning accuracy deteriorate when the proportion of dynamic objects is large
Solution Approach 1:
The patent extracts and removes feature points corresponding to dynamic objects from the feature set used for SLAM positioning. By identifying and eliminating these harmful feature points through projection position comparison and optical flow analysis, the system prevents dynamic objects from interfering with positioning accuracy while maintaining the benefits of visual SLAM.
2Quantity of substance
If feature points from all image regions are used for positioning, then the positioning information is abundant, but the positioning stability deteriorates due to inclusion of dynamic objects
Solution Approach 1:
The patent applies local quality by treating different feature points differently based on their spatial characteristics. Feature points are categorized into static and dynamic types based on their projection position consistency across frames. Static feature points are retained for positioning while dynamic feature points are removed, creating a heterogeneous feature set with optimized local properties for each region.
3Measurement precision
If optical flow algorithm is used to filter dynamic objects, then the positioning accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary action by pre-calculating and storing the projection positions of feature points from the previous frame before processing the current frame. This preprocessing step enables efficient comparison and identification of dynamic objects using optical flow, reducing the computational burden during real-time positioning by organizing data in advance.
Data Source
AI summary
The present disclosure provides a robot visual image feature extraction method as well as an apparatus and a robot using the same. The method includes: collecting image data through visual sensor(s) of the robot, and collecting angular velocity data through inertial sensor(s) of the robot; calculating a relative pose between image frames in the image data based on the angular velocity data; extracting feature points of the first image frame in the image data; calculating a projection position of each feature point of the k-th image frame in the k+1-th image frame based on a relative pose between the k-th image frame and the k+1-th image frame; and searching for each feature point in the projection position in the k+1-th image frame, and performing a synchronous positioning and a mapping based on the searched feature point. In this manner, the feature points of dynamic objects are eliminated.