摘要
现有的基于深度学习的视觉里程计(visual odometry,VO)训练样本与应用场景存在差异时,普遍存在难以适应新环境的问题,因此提出了一种在线更新单目视觉里程计算法OUMVO。其特点在于应用阶段利用实时采集到的图像序列在线优化位姿估计网络模型,提高网络的泛化能力和对新环境的适用能力。该方法使用了自监督学习方法,无须额外标注地面真值,并采用了Transformer对图像流进行序列建模,以充分利用局部窗口内的视觉信息,提高位姿估计精度,以避免传统方法只能利用相邻两帧图像来估计位姿的局限,还可以弥补采用RNN进行序列建模无法并行计算的缺点。此外,采用图像空间几何一致性约束,解决了传统单目视觉里程计算法存在的尺度漂移问题。在KITTI数据集上的定量和定性实验结果表明,OUMVO的位姿估计精度和对新环境的适应能力均优于现有的先进单目视觉里程计方法。
When training samples of existing deep learning-based visual odometry(VO)are different from application scena-rios,it is difficult to adapt to the new environment.Therefore,this paper proposed an online updated monocular visual mileage calculation method(OUMVO).In the application stage,it optimized the pose estimation network model online by using the real-time image sequence,which improved the generalization ability of the network and the ability to apply to the new environment.At the same time,it utilized self-supervised learning method without the need to mark the ground truth.Moreover,it adopted Transformer to conduct sequential modeling of image streams to make full use of the visual information within the local window to improve the precision of the pose estimation in order to avoid the limitation that the traditional method could only use two adjacent frames to estimate the pose.It could also compensate for the shortcomings of using RNN for sequence modeling which could not be calculated in parallel.In addition,it used the geometric consistency constraint of the image space to solve the scale drift problem of the traditional monocular visual mileage calculation method.Quantitative and qualitative experimental results on the KITTI dataset show that the proposed method is superior to existing state-of-the-art monocular visual odometry methods in terms of pose estimation accuracy and adaptability to new environments.
作者
王铭敏
佃松宜
钟羽中
Wang Mingmin;Dian Songyi;Zhong Yuzhong(School of Electrical Engineering,Sichuan University,Chengdu 610065,China)
出处
《计算机应用研究》
CSCD
北大核心
2024年第7期2209-2214,共6页
Application Research of Computers
基金
国家重点研发计划资助项目(2018YFB1307402)。
