基于改进YOLOv3和卡尔曼滤波器的飞机检测追踪方法

Aircraft Detection and Tracking Method Based on Improved YOLOv3 and Kalman Filter

精确的飞机检测与追踪方法有助于提升我国军事实力,但是目前针对小目标飞机进行有效追踪的方法较少。基于深度学习的目标追踪方法较传统的方法性能更佳优越,因此针对传统方法对于小目标追踪性能不佳的问题,提出了一种基于YOLOv3以及卡尔曼滤波器的飞机追踪方法,以获得更好的追踪性能。该算法首先通过改进的YOLOv3算法对视频中的图像进行检测,在识别到视频中的飞机之后,通过卡尔曼滤波器对飞机的运动轨迹进行预测,并通过匈牙利算法进行数据关联。实验结果显示,该算法对小尺度飞机的检测性能较传统的YOLOv3有接近5%的提升,且对飞机的追踪效果精度高且实时性好,具有较高的军事应用价值。

Accurate aircraft detection and tracking methods can help improve a country’s military strength, but there are few effective tracking methods for small target aircraft. The object tracking methods based on deep learning have better performance than traditional methods. Therefore, in view of the poor performance of traditional methods for small target tracking, this paper proposes an aircraft tracking method based on YOLOv3 and Kalman filter to obtain better performance. The improved algorithm first detects the target through the improved YOLOv3 algorithm, after that, the Kalman filter is used to predict the aircraft’s trajectory, and the Hungarian algorithm is used for data association. Experimental results show that the detection performance of the algorithm for small-scale aircraft is improved by nearly 5% compared with the traditional YOLOv3, and the tracking effect of the aircraft has high accuracy and real-time performance, which has high military application value.