基于视锥的多模态融合目标检测算法

Multi-modal Fusion Target Detection Algorithm Based on Frustum

三维目标检测准确且轻量化,是移动智能机器人感知模块的关键部分。本文提出一种基于视锥的多模态融合目标检测算法F-ResPoint。它是一种轻量化的二阶段目标检测算法,是根据二维目标的检测结果,生成包含目标对象的视锥点云,再利用基于Residual-SA模块构建的点云检测网络,对视锥点云中的三维目标进行检测。实验表明:该方法在轻量化的前提下,可实现较高的检测准确率及较好的实时性。

Three-dimensional object detection is accurate and lightweight, which is a key part of the perception module of intelligent mobile robots.In this paper, a cone-based multi-mode fusion target detection algorithm F-ResPoint is proposed.It is a lightweight two-stage target detection algorithm, which generates cone point cloud containing the target object according to the detection results of two-dimensional targets, and then uses the point cloud detection network built based on Residual-SA module to detect the three-dimensional targets in the cone point cloud.Experimental results show that this method can achieve higher detection accuracy and better real-time performance under the premise of lightweight.