基于深度学习的单目视觉水下机器人目标跟踪方法研究

介绍了一种基于深度学习的单目视觉水下机器人目标跟踪新方法,采用该方法能够使自主水下机器人(AUV)在水下复杂环境中实现图像增强处理与目标识别跟踪,对于视频图像中的每个图像,使用预先经过训练的深度卷积神经网络计算图像传输图,图像传输图提供了图像深度信息的相关估计,该方法能够识别水下目标区域,并标明水下目标运动及跟踪方向。实验表明:该方法能够更精确、更稳定地获取水下环境中的定位数据。

Research on Monocular Vision Target Tracking Method for Underwater Robot Based on Deep Learning

A new method of target tracking for monocular vision underwater robot based on deep learning is introduced. This method could be adopted to realize autonomous underwater vehicle (AUV) image enhancement processing and target recognition tracking in underwater complex environment for each acquisition. Every video frame was calculated using a pre-processed deep convolutional neural network. A correlation estimate of the image depth information was provided by the image transmission map. The underwater target area could be identified by the method and the underwater target motion and tracking direction be indicated. The experiment shows that the method can obtain the positioning data in the underwater environment more accurately and stably.