基于区域卷积神经网络和光流法的目标跟踪

Object Tracking Based on Region Convolution Neural Network and Optical Flow Method

为解决基于深度学习的在线目标跟踪算法速度慢的问题,设计并实现了一种基于区域卷积网络和光流法相结合的目标跟踪算法。该算法在T-1帧跟踪结果的基础上使用光流法计算跟踪目标的运动矢量计算出跟踪目标在T帧上的初选框,再将初选框区域作为区域卷积网络的输入,计算目标的精确跟踪结果。通过实验分析对比,算法对目标运动速度和形变具有很好的鲁棒性,并且跟踪速度可以达到50 frame/s。相较于在线跟踪算法,所提方法在满足较高的跟踪准确率的基础上大大提升了目标跟踪算法的速度。

In order to solve the problem of slow speed of online target tracking based on deep learning, a target tracking algorithm based on regional convolutional network and optical flow method is designed and implemented. Based on the T-1 frame tracking results, the optical flow method is used to calculate the tracking target＇s motion vector to calculate the primary selection box of the tracking target on the T frame, and the primary selection box is used as the input of the regional convolutional network to calculate accu-rate tracking of the target results. Through experimental analysis and comparison, the algorithm has good robustness to the target velocity and deformation,and the tracking speed can reach 50 frame/s. Compared with the on-line tracking algorithm, the proposed algorithm improves the speed of target tracking algorithm greatly while satisfying the high tracking accuracy.