基于图像弱检测器的飞机起降过程追踪方法

Landing and taking-off airplane tracking with weak detectors

在视频自动追踪拍摄飞机起降过程的应用环境中,面对高实时性和可靠性要求,存在目标成像变化大、背景复杂且高速运动、同类目标干扰等技术困难。针对以上问题,提出了一种仅通过几个图像弱检测器的组合,结合飞机4D航迹估算,实现稳定可靠的飞机实时追踪,并有效避免同类目标干扰的追踪方法。首先,将飞机起降过程分为几个阶段,每个阶段分别独立训练一个快速的弱检测器;然后,从单张2D图像中估算3D空间位置;最后,融合多张图像航迹估算信息结合航迹经验信息,对航迹进行精确计算,并预测下一时刻航班位置,最终实现稳定追踪。与其他追踪方法相比,新方法具有运算速度快、抗同类目标干扰能力强的优势。跟踪拍摄教练机起降训练过程的大量实验结果显示,该方法可以实现稳定可靠的飞机起降过程自动检测追踪。

For video tracking of landing and taking-off airplanes in real-time, the algorithm must be fast and reliable, and it must be capable of solving technical challenges, such as high degree of appearance feature varying, moving complex background, inference from other airplanes. An approach based on a combination of a few weak cascade detectors together with a 4D trace estimation algorithm was proposed. Firstly, the landing and taking-off procedure was divided into five stages. A weak cascade detector was trained for each stage. Then, a 3D position of airplane was estimated from a 2D image. Finally, the 4D trace was revised and the 3D position for the next time step was predicted based on the result of infommtion fusion from image detection and prior information on statistics of traces. This approach was tested on-site to track landing and taking-off coach airplanes. The experimental results show that this approach is able to track airplanes stably in real-time.