Driverless Vehicle Tracking Algorithm Based on Convolutional Neural network

To improve the accuracy and stability of tracking, this paper proposes an improved convolutional neural network (CNN) for driverless vehicle tracking. Firstly, a decaying learning rate is used in the optimizer by introducing learning rate decay technology. Then, we replace the convolution layers in the convolution blocks with the identity blocks to achieve residual nesting, which can better learn the complex feature information. The improved convolution blocks are used to construct residual network. Finally, experiments and comparisons show that the validation loss of the proposed CNN is significantly lower than that of the conventional CNN, which demonstrates the superiority of the proposed tracking method.