Int J Performability Eng ›› 2020, Vol. 16 ›› Issue (5): 720-727.doi: 10.23940/ijpe.20.05.p5.720727

• Orginal Article • Previous Articles     Next Articles

Visual Tracking Method based on Monte Carlo Compressed Sensing

Zirong Honga*(),  and Bo Danb   

  1. aSchool of Electric and Electronic Engineering, Lanzhou Petrochemical Polytechnic, Lanzhou, 730060, China
    bNaval Aviation University, Yantai, 264001, China
  • Submitted on ; Revised on ; Accepted on
  • Contact: Zirong Hong

Abstract: Aim

ing at the shortcomings of the existing compressed sensing visual tracking method, which has obstacles and lacks an updating mechanism for sampling templates, a Monte Carlo compressed sensing based robot visual tracking method is proposed. A small number of random particles relative to the original image pixel points are used to reduce the dimension greatly and extract the target features. At the same time, the sampled particles are updated every frame, highlighting the positive samples, suppressing the negative samples, and avoiding the blindness of random matrix sampling. The experimental results show that the new algorithm enables low complexity and accurate perceptual tracking of robot vision targets, overcomes the defects that the traditional tracking process is easily interfered by obstacles or target motion, and improves the tracking reliability.

Key words: visual tracking, compressed sensing, Monte Carlo sampling, robot