Automatic driving system and method based on monocular camera and raspberry pie

Abstract

The invention relates to an automatic driving system and method based on a monocular camera and the raspberry pie. The automatic driving system comprises a data collecting unit, a data preprocessing unit, a deep convolutional neural network and a control unit connected in sequence. The data collecting unit is used for collecting a data set. The data preprocessing unit is used for preprocessing thecollected data set. The deep convolutional neural network is used for training the preprocessed data set to obtain a mature model. The control unit is used for training the obtained mature model to enable the model vehicle to run automatically on a model lane. According to the invention, the expensive hardware like the complicated deep network model and radar are abandoned; automatic driving under conditions of low computing power and poor hardware condition is realized; and thus the unmanned driving is realized by using an end-to-end neural network and relying on the pure visual recognitionwith low costs while only the monocular camera and the raspberry pie are used.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an automatic driving system and method based on a deep convolutional neural network of a monocular camera running on a raspberry pie. Background technique [0002] At present, the rapid development of autonomous driving technology is generally based on multi-eye cameras or lidar, and needs to be run on high-configuration servers. However, the traditional Raspberry Pi provides less computing power and cannot provide traditional driverless technology The required computing power, when the hardware is not rich, that is, only one monocular camera is provided, because the monocular camera cannot provide more detailed road condition information when the model car is driving, so it is necessary to realize automatic driving A more complex neural network is required to process images captured by a monocular camera. However, as mentioned above, the computing power prov...

AI Technical Summary

Problems solved by technology

[0002] At present, the rapid development of autonomous driving technology is generally based on multi-eye cameras or lidar, and needs to be run on high-configuration servers. However, the traditional Raspberry Pi provides less computing power and cannot provide traditional driverless technology The required computing power, when the hardware is not rich, that is, only one monocular camera is provided, because the monocular camera cannot provide more detailed road condition information when the model car is driving, so it is necessary to realize automatic driving A more complex neural network is required to process images captured by a monocular camera. However, as mentioned above, the computing power provided by the Raspberry Pi is limited, so more complex neural networks cannot run on the Raspberry Pi. At present, there is no suitable neural network model that can achieve the balance between the two. In the existing network models, binocular cameras or trinocular cameras are generally used, and radar and other auxiliary equipment are used to assist simple multi-layer convolutional neural networks. To realize the automatic driving on the computing power of the Raspberry Pi, but this increases the hardware cost, so the automatic driving cannot be realized at this stage based on the hardware of the monocular camera and the Raspberry Pi

Images