Recognition of Motion Blurred Coded Targets Based on Convolutional Neural Network

In order to recognize the motion blurred coded targets in 3D vision measurement of moving objects, a method based on the convolutional neural network was proposed. First, a carefully designed convolutional neural network（MBCNet） was constructed and analyzed. Second, by exploiting the motion blur formation mechanism, a motion blurred image simulating system driven by six parameters was designed, so as to provide the large amount of motion blurred images of the coded targets needed for the network training. Up to 665 000 simulated motion blurred images of 100 kinds of coded targets were automatically generated and used for training the network. The recognition accuracy of 15 000 real motion blurred images of 5 kinds of coded targets reached up to 92.51%, which shows that the method for simulating motion blurred images is effective and the proposed network MBCNet has good generalization performance.