Optimization Design Of Intelligent Vehicle Obstacle Avoidance Based On Deep Learning

The intelligent car integrates advanced technologies such as image processing,intelligent control and radar positioning.It is widely used in various fields such as unmanned driving,emergency rescue,regional exploration,industrial production and family life.In recent years,with the continuous development of neural network theory,artificial intelligence has been paid more attention in the academic and industrial circles.The deep learning technology originated from artificial neural network has also made a vigorous development,especially in the field of image processing,which brings image processing technology to a new height.To the research of driverless technology,tracking and obstacle avoidance ability isthe most basic and important function of intelligent vehicle.In this paper,edge detection algorithm and threshold segmentation method are used for the realization of tracking function;The convolutional neural network algorithm is used in the implementation of the obstacle avoidance function,and then the algorithm is improved and verified on the intelligent car platform based on Raspberry PI.The main contents and emphases of this paper are as follows:1.The convolution neural network model of double convolution layer and doubleconnection layer are designed.This paper sets the step length of convolution kernel to 2,so that the pooling layer is eliminated.As a result,the scale of model is reduced,and the calculation speed of model is accelerated.The improved Adam optimizer is designed to solve the problem of poor convergence.The data loss of the improved convolutional neural network model is reduced by 3.3%and the accuracy is improved by 3.4%.2.The intelligent car experiment platform based on raspberry PI with a system built-in Linux system is built.The program is programed with Python 3 compiler development environment.The wireless serial port module is used to realize the wireless control of the smart car.The image data collection and preprocessing use OpenCV,which solves the problems of image distortion and the like.The intelligent PID control is used as the solution of the lateral control problem of the intelligent car to realize the PID self-adaptive adjustment.3.The image data is collected with the built intelligent car platform and grayed,filtered and clipped by OpenCV.The processed image data are transferred to the designed convolution neural network model and trained model,and the trained model is used to control the intelligent car for runway obstacle avoidance test.The obstacle avoidance and runway driving task can be completed by running on the intelligent car platform.The results show that the accuracy of the convolution neural network of the without-pooling layer is not affected.The improved Adam optimizer reduces the data loss in the training process.The obstacle avoidance and tracking task can be completed by running on the intelligent car with the improved model.