摘要
单目视觉进行目标识别有着巨大优势,但在目标测距方面存在精度不足且测量过程不稳定的问题,一种基于4线激光雷达与摄像头融合的联合测距的方法被提出并改善这一问题。首先利用卷积神经网络检测图像中的目标,得到相应的检测框;与此同时,通过标定相机内外参,将三维的激光点云数据转换到二维平面,得到2种数据对于检测环境的一致性表达。然后利用R-Tree算法快速配准检测框与相应的点云数据。此时,利用点云的深度信息能获得目标在真实世界的位置,并提出联合测距的方法来进一步提高测距精度。最终经实车采集的数据验证了所提算法的有效性。
Monocular vision has excellent performance in objects detection,but poor in distance measurement in terms of precise and robustness.A method named Joint Ranging was proposed,which bases on data fusion of 4 lines laser-scanner and camera,to improve that.Firstly,we use Convolutional Neural Networks(CNN)to detect image.Meanwhile,the lidar data obtained from environment need to map to image through a space transfer matrix.Then,the R-Tree algorithm is used to quickly match the lidar data and the corresponding detection boxes.Finally,the real location of objects could be calculated easily from fusion data.The joint ranging method was proposed to improve the precise and robustness,which was verified by dataset collected from real scenes.
作者
胡远志
刘俊生
肖佐仁
耿庄程
HU Yuanzhi;LIU Junsheng;XIAO Zuoren;GENG Zhuangcheng(State Key Laboratory of Vehicle NVH and Safety Technology,Chongqing University of Technology,Chongqing 400054,China;.Key Laboratory of Advanced Manufacturing Technology for Automobile Parts,Chongqing University of Technology,Chongqing 400054,China)
出处
《重庆理工大学学报（自然科学）》
CAS
北大核心
2019年第12期18-25,共8页
Journal of Chongqing University of Technology：Natural Science
基金
汽车噪声振动和安全技术国家重点实验室2019年度开放基金资助项目(NVHSKL-201908)
