Target tracking algorithm combining hybrid feature matching with Viterbi data association
Yanhua XU,Rong LI,Huajun WANG,Pingping XU
DOI: https://doi.org/10.16652/j.issn.1004-373x.2016.17.001
2016-01-01
Abstract:The tracking effect of the traditional tracking algorithms is poor due to the low video resolution,blurred frame image or heavy noise. To solve the above problems,a target tracking algorithm combining hybrid feature matching with Viterbi data association is proposed. The histogram back?projection technology is used to effectively segment the target edge in local bi?threshold image to overcome the problem of low contrast. The neighborhood feature,regional feature,movement direction feature and histogram feature are taken as the target characterization features to establish the cost function of hybrid feature. The Viterbi data association is used to calculate the sum of cost function to obtain the most similar target. The experimental results show that,in the condition of blurred frame image or heavy noise,the proposed algorithm has stable and effective target tracking,the accuracy rate of single target tracking is 0.89,the accuracy rate of multi?target tracking is 0.975,the recall rate is 0.920,and the algorithm is superior to other similar tracking algorithms.