计算机应用

• 人工智能与仿真 •    下一篇

空中加油场景下的目标联合检测跟踪算法

张怡1,孙永荣2,赵科东1,李华1,曾庆化1   

  1. 1. 南京航空航天大学自动化学院导航研究中心
    2. 南京航空航天大学
  • 收稿日期:2021-07-16 修回日期:2021-09-07 发布日期:2021-09-27 出版日期:2021-09-27
  • 通讯作者: 张怡

Joint detection and tracking algorithm of target in aerial refueling

  • Received:2021-07-16 Revised:2021-09-07 Online:2021-09-27 Published:2021-09-27

摘要: 摘 要: 针对自主空中加油对接阶段的目标跟踪问题,提出一种空中加油场景下的目标联合检测跟踪算法,采用检测跟踪一体化的CenterTrack网络实现对锥套的追踪,分别从模型设计与网络优化两方面改善该网络计算量较大、训练耗时过长的问题。首先在跟踪器中引入膨胀卷积组,在不改变感受野大小的前提下使得网络轻量化,同时将输出部分的卷积层替换为深度可分离卷积层,减少网络的参数量与计算量,然后对网络进行进一步的优化,将随机梯度下降法(SGD)与自适应矩估计算法(Adam)相结合,使其更快收敛至稳定状态。最后,利用真实的空中加油场景视频与地面模拟视频制作相应格式的数据集,供实验进行验证。最终分别在自制的锥套数据集和MOT17公共数据集上进行了训练与测试,算法的有效性得到了证实,改进的网络Tiny-CenterTrack减少了约48.6%的训练时长,并在实时性方面提升了8.8%。实验结果表明,改进后的网络在不损失网络性能的前提下可有效节省计算资源并在一定程度上提升实时性。

Abstract: Abstract: Focused on the issue of target tracking in the stage of autonomous aerial refueling, a joint detection and tracking algorithm of target in aerial refueling was proposed. CenterTrack network with integrated detection and tracking was adopted to track the drogue. In view of the large amount of calculation and long training time, this network was improved from two aspects: model design and network optimization. Dilation convolution group was firstly introduced into the tracker to make the network lighter without changing the size of the receptive field. At the same time, the convolutional layer of the output part was replaced with deeply separable convolutional layer to reduce the amount of network parameters and calculations. Moreover, the network was further optimized to make it converge to a stable state faster by combining stochastic gradient descent(SGD) method with adaptive moment estimation (Adam). Finally, videos which shooted during aerial refueling process and simulated on the ground were made into corresponding format for experimental verification. Eventually, model were trained and tested respectively on the drogue dataset and MOT17(Multiple Object Tracking 17) dataset. The effectiveness of the algorithm has been verified. The improved network Tiny-CenterTrack reduces training time by about 48.6% and improves the real-time performance by 8.8%. Experimental results show that the improved network can effectively save computing resources and boost real-time performance to a certain extent without loss of network performance.

中图分类号: