Visual positioning system according to semantic meaning in dynamic environment and method thereof

Abstract

The invention discloses a monocular vision positioning system in a dynamic environment and a method thereof. According to semantic meaning, elimination of a dynamic object characteristic is realized.An environment image is acquired in real time through a monocular image camera. In the image, object detection is realized through a convolutional neural network, thereby obtaining the semantic meaning of the object. Furthermore determination of a dynamic object is realized according to priori knowledge. The characteristic in the image is extracted by means of an ORB algorithm, and furthermore thecharacteristic points of the dynamic object are eliminated according to the position of the dynamic object. Local bundle adjustment is performed on the gesture and the 3D point coordinate of the camera, thereby eliminating influence of the characteristic points of the dynamic object and improving positioning precision.

Description

technical field [0001] The invention relates to the field of computer vision and mobile robot positioning, in particular to a visual positioning system and method combined with semantics in a dynamic environment. Background technique [0002] Simultaneous Localization And Mapping (SLAM) is a method that uses sensor information to build an environmental map and determine its own pose. When using a camera as a sensor is called visual SLAM. Establishing a real-time, accurate and robust SLAM system is of great significance to the positioning of robots, unmanned vehicles and other equipment, and is the basis for realizing navigation and autonomous movement. [0003] Traditional SLAM technology is based on a static environment and does not consider the movement of environmental objects. In the actual environment, the movement of people and the movement of vehicles will cause dynamic changes in the environment, so that the map established by the SLAM system cannot maintain consis...

AI Technical Summary

Problems solved by technology

[0008] (1) The feature points of dynamic objects to be deleted are determined according to the Mahalanobis distance calculated by the error model of the field flow method. Using a fixed threshold for objects with different motion forms and speeds will lead to an increase in error and cannot correctly determine the object The dynamic characteristics of

[0009] (2) The error model of the field flow method makes a static background assumption, which only has the ability to detect moving objects in a small area, and cannot handle the situation where large moving objects appear in the field of view.

[0010] (3) Only objects that move between adjacent frames are considered as dynamic objects, without considering the dynamic characteristics of the object itself

Images