Aiming at the problem that excessive relative motion between target and camera during the exposure time will result in blurred imaging

which affects precision of the SLAM. On the basis of traditional motion deblurring method

a visual SLAM based on GAN is proposed. First of all

the GAN based on improved FPN is designed

which generates the sharp images from the blurred images end-to-end

and then the motion deblurring visual SLAM is acquired by combining the GAN with ORB-SLAM2. The method is carried out on the public dataset that tests the effect of deblurring

test results show that the method has better deblurring effect than similar algorithms

and the processing time is shorter. In order to verify the effect of the SLAM

the test is carried out on the public dataset EuRo C and the result shows that the total RMSE of the method is 6. 83 cm.