Hardware implementation method and system for real-time extraction and selection of video stream feature points

Abstract

The invention discloses a hardware implementation method and system for real-time extraction and selection of video stream feature points, and aims at effectively improving the speed of an algorithm by simplifying combinational logic on a critical path and designing a four-point parallel processing and pipeline structure; in order to solve the problems of poor real-time performance and low precision in the feature point selection process, by optimizing an image cache structure and designing a feature point register and a block selection and storage structure, the calculation amount is effectively reduced, and the precision and speed of an algorithm are improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a hardware implementation method and system for real-time extraction and selection of feature points of a video stream. Background technique [0002] Simultaneous localization and mapping (SLAM, Simultaneous Localization and Mapping), which refers to a mobile platform equipped with specific sensors, without current environmental information, constructs an environmental map model in the process of movement, and estimates its own positioning at the same time. As the premise of autonomous navigation tasks for mobile robots, this technology has always been a research hotspot in the field of robotics. Among them, visual SLAM using cameras as the main sensor will have broad application prospects in the future due to its advantages of low cost and rich information. [0003] In visual SLAM, the feature point processing part in the front-end visual odometry is a cruci...

AI Technical Summary

Problems solved by technology

However, this method will result in more features extracted from areas with rich textures, and no feature points in areas with weak textures. The uneven distribution of feature points in space will easily lead to an increase in the error of pose estimation, making this method more accurate. Poor, not suitable for engineering applications in actual scenarios

[0005] For this reason, for the problem of uneven selection of feature points, another method uses quadruple and sorting to achieve uniform feature point selection, but this introduces multiple uncertain iterative calculations, and the subsequent complex logic branches And repeated memory reading, resulting in poor real-time performance, unable to better complete the real-time processing tasks of high-speed video streams

Images