A Cross-Channel Attention-Enhanced 3D Vision-Based Point Cloud Place Recognition Model for Warehousing Environments

Point cloud based retrieval for place recognition is still a challenging problem due to pronounced variations in scene appearance across different environmental contexts. Addressing the limitations of existing point cloud-based place recognition methods in complex warehousing environments, which involve downsampling and inefficient utilization of point cloud input data, resulting in information loss. This paper introduces a novel place recognition network model based on Kernel Point Convolution (KPConv). This approach leverages both scene intensity and point cloud data as initial inputs, employing Kernel Point Convolution for feature extraction. Furthermore, a Cross-Channel Attention module is employed to enhance the network’s attention to overlapping parts of point clouds. Through comprehensive comparative analysis, our proposed network significantly enhances place recognition accuracy, demonstrating its efficacy in addressing the aforementioned shortcomings.