A Novel Dynamic Target Tracking Algorithm for Image Based on Two-step Reinforcement Learning

In this article, we modeled image target tracking into reinforcement learning framework, and we proposed a two-step reinforcement learning algorithm for target tracking. In this algorithm, we set multiple tracker agent to track the pixel of target, the intention of reinforcement learning is to achieve tracking strategy of every tracker agent, we divided each learning step of tracker into two parts, one is to learn the division strategy, another is to learn the action strategy, every tracker agent shares the experiences they have learned. Simulation experimental results illustrate the feasibility and effectiveness of the algorithm. Copyright © 2014 IFSA Publishing, S. L.