3PCD-TP: A 3D Point Cloud Descriptor for Loop Closure Detection with Twice Projection

Loop closure detection (LCD) can effectively eliminate the cumulative errors in simultaneous localization and mapping (SLAM) by detecting position of a revisit building interframe pose constraint relations. However, real-world natural scenes, driverless ground vehicles or robots usually same place from different position, meaning that descriptor cannot give uniform description similar failing LCD. Against this problem, paper proposes 3D point cloud with Twice Projection (3PCD-TP) for calculation similarities between scenes. First, we redefined origin primary direction clouds according to their distribution unified coordinate system, thereby reducing interference recognition due rotation translation sensors. Next, using semantic altitudinal information clouds, generated 3PCD-TP multidimensional features enhance its ability describe Following this, designed weighting similarity method reduce false rate LCD taking advantage property be projected multiple angles. Finally, validated our KITTI Jilin University (JLU) campus dataset. The experimental results show demonstrated high level precision recall exhibited greater performance face scenes reverse loop closure, such as opposite lanes.