View source for 17378155. METHODS AND SYSTEMS FOR SEMANTIC SEGMENTATION OF A POINT CLOUD simplified abstract (HUAWEI TECHNOLOGIES CO., LTD.)

=METHODS AND SYSTEMS FOR SEMANTIC SEGMENTATION OF A POINT CLOUD=


==Organization Name==

[[:Category:HUAWEI TECHNOLOGIES CO., LTD.|HUAWEI TECHNOLOGIES CO., LTD.]]


[[Category:HUAWEI TECHNOLOGIES CO., LTD.]]

==Inventor(s)==

[[:Category:Ran Cheng of Markham (CA)|Ran Cheng of Markham (CA)]][[Category:Ran Cheng of Markham (CA)]]

[[:Category:Ryan Razani of North York (CA)|Ryan Razani of North York (CA)]][[Category:Ryan Razani of North York (CA)]]

[[:Category:Bingbing Liu of Markham (CA)|Bingbing Liu of Markham (CA)]][[Category:Bingbing Liu of Markham (CA)]]

==METHODS AND SYSTEMS FOR SEMANTIC SEGMENTATION OF A POINT CLOUD - A simplified explanation of the abstract==


This abstract first appeared for US patent application 17378155 titled 'METHODS AND SYSTEMS FOR SEMANTIC SEGMENTATION OF A POINT CLOUD

==Simplified Explanation==

The patent application describes a system for segmenting 3D point clouds using deep neural networks. The system consists of two main components: a multi-branch cascaded subnetwork and a spatial feature transformer subnetwork.

* The multi-branch cascaded subnetwork includes an encoder and a decoder, which work together to process a sparse 3D point cloud. This subnetwork captures and combines spatial feature information from the point cloud at different scales and hierarchical levels.

* The spatial feature transformer subnetwork takes the features generated by the multi-branch cascaded subnetwork and transforms them. It then uses a shared decoder attention framework to fuse these transformed features and assist in predicting semantic classes for the sparse 3D point cloud.

Potential applications of this technology:

* Autonomous driving: The system can be used to segment 3D point clouds captured by LiDAR sensors in autonomous vehicles, helping to identify and classify objects in the environment.

* Robotics: The system can be applied to robotic systems that use 3D point cloud data for perception tasks, such as object recognition and scene understanding.

* Augmented reality: By segmenting 3D point clouds, the system can enhance the accuracy and realism of augmented reality applications, allowing virtual objects to interact more seamlessly with the real world.

Problems solved by this technology:

* Efficient segmentation: The system addresses the challenge of accurately segmenting 3D point clouds, which are sparse and contain complex spatial information. The multi-branch cascaded subnetwork and spatial feature transformer subnetwork work together to capture and fuse this information effectively.

* Semantic classification: The system helps to classify the different parts of a 3D point cloud into semantic classes, such as buildings, roads, or vegetation. This enables better understanding and interpretation of the point cloud data.

Benefits of this technology:

* Improved accuracy: By utilizing deep neural networks and multi-scale feature fusion, the system can achieve more accurate segmentation of 3D point clouds compared to traditional methods.

* Real-time processing: The system is designed to process point cloud data efficiently, making it suitable for real-time applications such as autonomous driving or robotics.

* Generalizability: The system's architecture and framework can be adapted and applied to various domains and tasks involving 3D point cloud segmentation.


==Original Abstract Submitted==

Systems, methods and apparatus for sematic segmentation of 3D point clouds using deep neural networks. The deep neural network generally has two primary subsystems: a multi-branch cascaded subnetwork that includes an encoder and a decoder, and is configured to receive a sparse 3D point cloud, and capture and fuse spatial feature information in the sparse 3D point cloud at multiple scales and multi hierarchical levels; and a spatial feature transformer subnetwork that is configured to transform the cascaded features generated by the multi-branch cascaded subnetwork and fuse these scaled features using a shared decoder attention framework to assist in the prediction of sematic classes for the sparse 3D point cloud.



[[Category:G06T7/11]]
[[Category:G06K9/62]]
[[Category:G01S17/89]]
[[Category:G01S17/931]]