基于RGB图像的坦克损伤目标三维检测研究与应用

Research and Application of Tank Damage Target 3D Detection Based on RGB Image

现代战争中,坦克在攻坚战中的地位越来越重要,检测坦克的损伤对于取得战场主动权乃至获取战争的胜利起着决定性作用,所以对实时性要求非常高。采用易获取的RGB图像,以坦克装甲车为研究目标,选用Complex-YOLO为基础三维目标检测模型,针对复杂战场环境中图像内容复杂、弹孔损伤目标小、没有三维CAD模型等问题,对Complex-YOLO模型进行改进,通过使用识别精度高且速度快的YOLOV3网络及九点法回归三维目标检测框的方法,提高模型性能。在坦克数据集上的实验结果表明,改进后的算法对于复杂战场环境下的多目标检测具有更强的敏感性,较大程度上增强了模型的检测识别精度。

In modern war,tanks play an increasingly important role in the assault of fortified positions.The detection of tank damage plays a decisive role in gaining the initiative in the battlefield and even winning the war,so the real-time requirement is very high.The RGB image is used which is easy to obtain,the tank is regarded as the research object,and the complex Yolo is selected as the basic three-dimensional target detection model.Aiming at the problems such as the complex content of the image in the complex battle field environment,small bullet hole damage target,no three-dimensional CAD model and so on,the complex Yolo model is improved by using the high identification precision and fast processing Yoov3 network and nine-point method to regress the three-dimensional image detection frame in order to improve the performance of the model.The experimental results on data set of tank show that the improved algorithm has stronger sensitivity for multi-target detection in complex battlefield environment,and can greatly enhance the detection and recognition precision of the model.