一种基于孪生网络的舰船目标跟踪方法

Ship Target Tracking Method Based on SiameseNet

针对未来海战场目标的跟踪自动化和智能化的需求,提出了一种基于孪生网络的海上目标跟踪方法。近年来随着孪生网络在图像识别领域的广泛应用,因此将这项新技术应用于海上舰船目标具有极强的现实意义。采用基于全连接孪生网络的目标跟踪算法,对海上目标进行跟踪与定位。首先采集目标的视频数据集,然后利用数据集进行网络训练,利用训练所得模型可对海上目标进行跟踪与定位。由于保密相关问题,舰船目标数据集由等比缩小的航模代替进行试验。由测试结果可知,系统所采用的SiameseNet算法,准确率可达80%,且平均帧率可达60fps。实时性、准确性能够基本满足实际应用需求。为了对上述算法进行有效调试,以及下一步满足工程化需求,将结合该算法,研制算法配套的硬件平台。

With the continuous development of military technology,various countries have an urgent need for precision strike weapons.It is of great significance to carry out research on scene matching technology.General image matching real-time image and template image is in different weather,different time,different angles and different imaging equipment condition,so the image will exist between the obvious distortion,therefore this paper proposes a method of scene matching technology based on scale invariant.First,SURF feature extraction is used to register real-time images and template images,and then template matching is carried out for interested targets to achieve scene matching.The method in this paper is not only robust to various distortion and transformation,but also capable of template matching for different objects,so as to realize the real-time processing of scene matching.