A Correlation Filter Target Tracking Algorithm Based on Continuous Convolution Operator and Spatial-Temporal Regularization Term

一种连续卷积与时空正则项的相关滤波目标跟踪算法

School of Electronics and Information Engineering, Xi'an Technological University, Xi'an 710021, China

Received: 11 January 2019

Abstract

In order to accurately estimate the location of a target that is occluded and rotated rapidly by the STRCF algorithm, a correlation filter target tracking algorithm based on continuous convolution operator and spatial-temporal regularization term is proposed. The algorithm uses interpolation operators to transform a response function into a continuous function within a certain period, thus enhancing the accuracy of target location. Spatial-temporal regularization terms are added to the new model of correlation filter to ensure that it is similar to the model of the previous frame of image and that the algorithm is more robust. A fast multi-scale filter is used to update the scale, thus improving the computational efficiency. The experimental results show that the average overlap rate of the proposed algorithm can reach 73% and that the central position error is less than 8.2. The proposed algorithm can achieve a real-time and robust target tracking.

摘要

针对STRCF算法在目标被遮挡及快速旋转时定位不准确的问题，提出了一种连续卷积与时空正则项的相关滤波算法。该算法利用插值算子将响应函数在一定周期内转换为连续函数，提高了目标定位精度；加入时空正则项构建相关滤波模型，保证模型与上一帧模型相似，提高算法鲁棒性；采用快速多尺度滤波对尺度进行更新，提高运算效率。实验结果表明，所提算法的平均重叠率可达73%，中心位置误差低于8.2个像素，可以实现对目标实时和鲁棒性跟踪。