Methods of simulating 3D point cloud data to represent unstructured industrial environments

Automation in the manufacturing and logistics sectors have seen substantial growth in recent years with the integration and continual development of autonomous robotic systems, such as Autonomous Guided Vehicles (AGVs). A current limitation of these systems is their inability to correctly and rapidly detect Objects Of Interest (OOI) using 3D image capture systems. Training vision systems to identify and plan interactions with OOIs requires hours of captured footage under various conditions to encompass all possible environments.

In this study, we propose a method for simulating 3D point clouds with scalable surface noise to allow for the rapid production of test data for the development of 3D processing systems. Currently, there are limited options available for producing simulated point cloud data. Current methods consist of producing complete 3D objects defined by a number of surfaces and vertex properties. Here, a method is presented that is capable of producing surface geometries with a defined point density and surface noise produced by artificial randomness.For this simulation method, point generation is dimension driven based on a defined object origin offset from the world origin. Each surface of the simulated object is dimensioned by endpoint constraints and plane orientation is defined by surface angles relevant to the object origin axes. Gaussian point noise is generated in the Z direction to simulate noise artefacts typically found in Time of Flight 3D systems.Results to date have shown that this simulation method is highly effective at producing simple load models with varying degrees of noise. The results of this study show that the noise generated indicates a standard distribution, representative of typical Gaussian noise, as observed in real-life data. This verifies that the data produced is suitable for training and testing machine vision systems, by providing data that closely resembled real-world noise and artefacts. This is unique to this method as no other current simulation process presents a controlled and scalable way to reproduce noise in simulated data.