一种多层次贝叶斯雷视融合目标检测与识别方法-《计算机技术与发展》

文章信息/Info

Title:: Radar-vision Fusion Target Detection and Recognition Based on Bayesian Networks

作者:: 刘晓莉1; 王洪龙1; 贺鹏鸣2; 周盼盼1; 王建强2; 1. 甘肃省交通投资管理有限公司,甘肃兰州 730030;
2. 兰州交通大学交通运输学院,甘肃兰州 730070

Author(s):: LIU Xiao-li1; WANG Hong-long1; HE Peng-ming2; ZHOU Pan-pan1; WANG Jian-qiang2; 1. Gansu Communication Investment Management Co. ,Ltd. ,Lanzhou 730030,China;
2. School of Traffic and Transportation,Lanzhou Jiaotong University,Lanzhou 730070,China

Keywords:: autonomous driving; target detection and recognition; Bayesian network; radar-vision fusion; parametric learning

摘要:: 随着自动驾驶技术的发展,精确的目标检测与识别成为确保行车安全的关键。现有的基于贝叶斯网络的目标检测方法大多局限于单一传感器数据,未能充分利用多传感器数据的互补性。该文提出了一种基于贝叶斯网络的雷视融合目标检测和识别方法,旨在通过融合雷达和视觉传感器数据提高检测的准确性和鲁棒性。该文构建了一个多层次贝叶斯网络模型,用于表示两种传感器数据间的依赖关系,并设计了一种新的参数学习算法来优化网络参数。此外,提出了一个基于网络推理的检测流程,实现了对目标的快速且准确检测。在标准数据集上的实验结果表明,与现有技术相比,该方法在多个评估指标上均有显著提升,包括准确率、召回率和 F1 分数,同时保持了较低的计算复杂度,能够实现实时的目标检测,适合实时处理需求。该研究还探讨了模型的泛化能力和潜在的应用场景,为自动驾驶车辆的感知系统提供了一种有效的解决方案。

Abstract:: With the development of autonomous driving technology,accurate target detection and recognition has become the key to ensure driving safety. Most of the existing target detection methods based on Bayesian networks are limited to single sensor data and fail to make full use of the complementarity of multi - sensor data. In this paper, a Bayesian network - based method for detection and recognition of lightning fusion targets is proposed to improve the accuracy and robustness of detection by fusing radar and vision sensor data. We construct a multi-level Bayesian network model to represent the dependency between the two sensor data,and design a new pa-rameter learning algorithm to optimize the network parameters. In addition,we propose a detection process based on network inference to achieve fast and accurate detection of the target. Experimental results on standard data sets show that compared with the prior art,the proposed method has significant improvement in multiple evaluation indicators, including accuracy, recall rate and F1 score, while maintaining low computational complexity, which can realize real - time target detection and is suitable for real - time processing requirements. This study also explores the generalization ability and potential application scenarios of the model,providing an effective solution for the perception system of autonomous vehicles.