Robot autonomous localization and navigation based on image-text recognition and semantic meaning

Abstract

The invention relates to autonomous localization and navigation of mobile intelligent equipment such as service robots, unmanned aerial vehicles, automatic guided vehicles and indoor three-dimensionalmodeling equipment. Compared with the existing system, image-text recognition and semantic binding for indoor symbols are introduced. The characteristics are that (1) the recognized result of image-text recognition is high in accuracy compared with laser radar or road signs for visual recognition; (2) only a camera is required to be added, and the cost and the weight are low; (3) an accumulativeerror of sensors (an IMU, laser radar, vision, etc.) is effectively eliminated, and the accuracy of loop detection is improved; (4) the robot localization and navigation accuracy is improved on the whole by combining image-text recognition, and commercial use of robots is enabled to become possible; and (5) tool software provided by using the method combines a result of image-text recognition, endows an environment map with semantic information through manual binding, can promote man-machine communication and helps the robots to complete advanced tasks.

Description

technical field

[0001] The present invention relates to the field of autonomous positioning and navigation of mobile smart devices such as service robots, unmanned aerial vehicles, automatic guided vehicles, and indoor 3D modeling equipment, and in particular to the positioning and navigation of smart mobile devices when there is no GPS signal indoors , this kind of positioning and navigation does not need to provide the floor plan of the environment in advance, and does not need to deploy wireless networks such as UWB and ZIGBEE networks indoors. Background technique

[0002] The demand for high-precision positioning comes from the booming and rapidly developing field of robots and wearable devices. For example, robots include housekeeping robots, sweeping robots, and logistics robots. Compared with human positioning, positioning is just needed for these devices, and The diversity of application scenarios cannot be realized by special equipment such as UWB base stations and...

Examples

Embodiment Construction

[0054] The present invention will be further described below in conjunction with embodiment.

[0055] The system designed by the invention is divided into an offline part and an online system.

[0056] Offline part:

[0057] It is mainly to complete the collection of various signs, especially indoor direction signs and warning signs, which are manually labeled (labelled) and stored in the database. Through the machine learning scheme, the algorithm memorizes the graphics, image features and features of these signs. semantic information. it includes:

[0058] 1) Collection: It can be collected on the network or manually on-site

[0059] 2) Labeling: Manually label the attributes and semantics of the collected signs, such as the shape and meaning of the direction signs, signs of passage and signs of prohibition, etc.

[0060] 3) Training: through machine learning methods, the algorithm can obtain the knowledge of these signs (attributes, concepts, semantics, etc.)

[0061] ...

technical field

[0001] The present invention relates to the field of autonomous positioning and navigation of mobile smart devices such as service robots, unmanned aerial vehicles, automatic guided vehicles, and indoor 3D modeling equipment, and in particular to the positioning and navigation of smart mobile devices when there is no GPS signal indoors , this kind of positioning and navigation does not need to provide the floor plan of the environment in advance, and does not need to deploy wireless networks such as UWB and ZIGBEE networks indoors. Background technique

[0002] The demand for high-precision positioning comes from the booming and rapidly developing field of robots and wearable devices. For example, robots include housekeeping robots, sweeping robots, and logistics robots. Compared with human positioning, positioning is just needed for these devices, and The diversity of application scenarios cannot be realized by special equipment such as UWB base stations and...

Examples

Embodiment Construction

[0054] The present invention will be further described below in conjunction with embodiment.

[0055] The system designed by the invention is divided into an offline part and an online system.

[0056] Offline part:

[0057] It is mainly to complete the collection of various signs, especially indoor direction signs and warning signs, which are manually labeled (labelled) and stored in the database. Through the machine learning scheme, the algorithm memorizes the graphics, image features and features of these signs. semantic information. it includes:

[0058] 1) Collection: It can be collected on the network or manually on-site

[0059] 2) Labeling: Manually label the attributes and semantics of the collected signs, such as the shape and meaning of the direction signs, signs of passage and signs of prohibition, etc.

[0060] 3) Training: through machine learning methods, the algorithm can obtain the knowledge of these signs (attributes, concepts, semantics, etc.)

[0061] ...

Images