Q. Is there a 2D Lidar model for neural net?

 

A.

Yes, there are many neural network models specifically designed for processing 2D Lidar data.

These models are typically used to perform either Semantic Segmentation (classifying points as 'wall', 'person', 'car', etc.) or Object Detection and Tracking (finding the bounding box and ID of a moving object), often outperforming traditional methods like DBSCAN and simple nearest-neighbor tracking in complex environments.

Here are the primary types of neural network models used with 2D Lidar data:

---

 

1. Grid-Based (Image-like) Models

 

The simplest way to use standard Convolutional Neural Networks (CNNs) with 2D Lidar is to convert the circular scan into a 2D image format.

 

 

 

 

 

For this project, moving to a PointNet-like approach would be the logical next step to replace the DBSCAN part of your pipeline, allowing the network to learn optimal ways to group points into objects.