Design And Implementation Of Driver Assistance System For Unmanned Vehicles Based On Convolutional Neural Network

In recent years,with the advent of the era of artificial intelligence,artificial intelligence technology has been widely used in all walks of life,among which the most popular is undoubtedly the unmanned technology that has been studied in recent years.With the development of driverless technology,all kinds of auxiliary driving systems on the car,as the key part of safe driving,can make the car carry out real-time warning,and actively intervene to avoid all kinds of accidents.Through the above analysis points,this paper proposes a kind of assistant driving system based on the high-performance embedded platform of Jetson nano,which is realized by using machine vision and deep learning technology.The system includes two parts: the classification of circular and triangular traffic signs,and the detection of pedestrians.The main contents of this paper include:(1)Aiming at the traffic sign module,this paper proposes an idea of segmentation first,then classification,that is,first simulate the car camera to take pictures,and then through the histogram equalization,sharpening,Retinex enhancement and other preprocessing operations,according to the unique HSV characteristics of traffic signs,detect the approximate area of color,and then use morphological operations to further denoise After that,the traditional Hof transform is used to detect the specific shape,and the traffic signs are segmented from the overall image,which is used as the input of the neural network for the classification and recognition of traffic signs.The neural network training part uses the GTSRB data set for training and data expansion on this basis.Lenet-5 convolution neural network is selected for the network,and the experimental analysis and verification are completed at last.(2)For the pedestrian recognition module,this paper analyzes the traditional detection methods,and uses an improved convolutional neural network yolov3 tiny as the training network,and uses labelimg annotation tool to create the data set,and analyzes the specific results of the training according to the training log.(3)According to the requirement analysis,a complete system structure is designed,the work flow of the system is analyzed,and the software and hardware environment are introduced.Then the classification experiment is carried out for the traffic sign module,and the detection accuracy is analyzed.Finally,on the platform of Jetson nano,the pedestrian detection experiments are carried out on the regular state,dynamic fuzzy state,long-distance and overlapping state.