基于激光雷达和相机融合的目标检测算法  

Research on Object Detection Algorithm based on Lidar and Camera Fusion

作　　者：李梦楠 徐晓光[1,2,3] LI Mengnan;XU Xiaoguang(Key Laboratory of Advanced Perception and Intelligent Control of High-end Equipment,Ministry of Education,Wuhu 241000,China;Anhui Polytechnic University,college of electrical engineering,Wuhu 241000,China;Anhui Polytechnic University,Industrial Innovation Technology Research Co Ltd,Wuhu 241000,China)

机构地区：[1]高端装备先进感知与智能控制教育部重点实验室,安徽芜湖241000 [2]安徽工程大学电气工程学院,安徽芜湖241000 [3]安徽工程大学产业创新技术研究有限公司,安徽芜湖241000

出　　处：《牡丹江师范学院学报（自然科学版）》2024年第4期17-21,共5页Journal of Mudanjiang Normal University：Natural Sciences Edition

基　　金：国家自然科学基金区域创新发展联合基金重点支持项目(U22A2079);安徽省高校自然科学研究项目(KJ2020A0350);安徽未来技术研究院企业合作项目(2023qyhz29);安徽工程大学-鸠江区产业协同创新专项基金项目(2022cyxtb9)。

摘　　要：基于多模态焦点稀疏卷积的目标检测算法对小尺寸目标检测精度不足的问题,提出一种多尺度特征提取网络.算法融合多尺度特征,提取网络优化原始算法中的特征提取部分,提升行人和自行车类别的定位精度,加入多模态焦点稀疏卷积目标检测网络.结果表明,改进后的目标检测模型提升了小目标“Pedestrians”类和“Cyclist”类检测的mAP(mean Average Precision)百分比,提升了小尺度目标的定位精度.In this paper,a multi-scale feature extraction network is proposed based on the problem that the target detection algorithm with multi-modal focus sparse convolution has insufficient detection accuracy for small-size targets.The algorithm fuses the multi-scale features,extracts the network to optimise the feature extraction part of the original algorithm,improves the positioning accuracy of pedestrian and bicycle categories,and incorporates a multimodal focal sparse convolution target detection network.The results show that the improved target detection model improves the mAP(mean Average Precision)percentage for detecting small targets of Pedestrians and Cyclists,and improves the localisation accuracy of small-scale targets.scale target localisation accuracy.

关 键 词：多尺度卷积网络 特征提取 点云 稀疏卷积 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]