Our paper entitled “DynamicFilter: an Online Dynamic Objects Removal Framework for Highly Dynamic Environments” is accepted by ICRA 2022.
Collecting data with dense crowds in the mall
The floor plan of the mall
The point cloud map built by our approach (The red points are the dynamic pedestrians to be removed. The white points are the spatial structure of the environment)
The emergence of massive dynamic objects will diversify spatial structures when robots navigate in urban environments. Therefore, online removal of dynamic objects is critical.
This paper proposes a novel online removal framework for highly dynamic urban environments. The framework consists of a scan-to-map front-end and a map-to-map back-end. Both two ends deeply integrate visibility-based approaches and map-based approaches. Especially, in the back-end, we present a visibility check that uses a visibility-based approach to approximate the ray-tracing process and accelerate its occupancy computation.
In the experiment, considering dynamic objects in the SemanticKITTI dataset does not appear frequently, we introduce a simulation environment with more dynamic objects. And we also tests our algorithm in real-world crowded environment. The experiment demonstrates the validity of the framework in real-world datasets and highly dynamic simulation scenarios.