Aiming at the problem of relocalization after robot abduction

a global localization algorithm based on particle filtering and image oriented FAST and rotated BRIEF（ORB） feature matching is proposed. When the problems occur

the set of particles predicted by the motion model cannot correctly estimate the pose of the robot.The particle set is modified by adding observation results of the camera. The location of the robot is detected based on ORB feature matching of the camera image. The particles are generated in the grid map area associated with the camera. The relocalization is achieved by observation update and re-sampling to gradually converge the particles.The experimental results show that the proposed algorithm can solve the kidnapped robot problem and is superior to the adaptive Monte Carlo localization（AMCL） algorithm with random particles in terms of time efficiency

and it has a lower positioning error.