张静雨 梁佳佳 汪文卉

摘 要：在传统的视觉寻迹智能车中，人们使用摄像头获得路况信息，并对图像加以处理，然后根据产生的偏差来控制小车行驶，它会根据设定好的判定条件实现方向控制。而人工训练的神经网络能够在训练中不断优化神经突触的权值，并将信息存储在内部，通过大量数据的学习，实现寻迹功能。本项目摒弃了传统的电磁采集位置式PID控制算法，使用自行训练的人工神经网络模型对电磁传感器的电磁进行分析。车模方向控制方面，不再手动编写控制算法和调节参数，使训练集、测试集等符合高斯分布。这样不仅让训练出的神经网络具有一定的鲁棒性，还能够增强智能车对道路的适应性。人工神经网络控制算法可以应用到更多的人工智能领域，人们可以类比基于机器视觉的神经网络控制模型，利用openCV（可编程全彩摄像头），配合神经网络算法，为无人驾驶技术在3D感知与物体检测中的应用打下坚实基础。同时，人们可以在其他领域做出进一步的尝试：电子导盲犬利用openCV，配合特定算法，能很好地捕捉和反馈路面信息;垃圾分类机器人利用openCV强大的图像处理库，配合卷积神经网络，可以实现垃圾的智能分类。

关键词：人工神经网络;智能车;模块化

中图分类号：TP18文献标识码：A文章编号：1003-5168（2021）04-0053-03

Abstract： In the traditional visual tracing smart car， people use the camera to obtain road condition information， process the image， and then control the driving of the car according to the deviation generated， and it will realize the direction control according to the set judgment conditions. The artificially trained neural network can continuously optimize the weight of the nerve synapse during training， and store the information internally， and realize the tracking function through the learning of a large amount of data. This project abandons the traditional electromagnetic acquisition position PID control algorithm， and uses a self-trained artificial neural network model to analyze the electromagnetic sensor's electromagnetic. Regarding the direction control of the car model， the control algorithm and adjustment parameters are no longer manually written， so that the training set， test set， etc. conform to the Gaussian distribution. This not only allows the trained neural network to have a certain degree of robustness， but also enhances the adaptability of the smart car to the road. Artificial neural network control algorithms can be applied to more artificial intelligence fields， people can compare neural network control models based on machine vision， use openCV （programmable full-color camera）， and neural network algorithms to lay a solid foundation for the application of driverless technology in 3D perception and object detection. At the same time， people can make further attempts in other areas： electronic guide dogs use openCV with specific algorithms to capture and feedback road information well; the garbage sorting robot uses the powerful image processing library of openCV and the convolutional neural network to realize the intelligent classification of garbage.

Keywords： artificial neural network;intelligent car;modularity

智能車是典型的高新技术综合体，其研究成果广泛应用于交通运输事业，是世界车辆工程领域研究的热点[1-2]。传统电磁采集位置式PID算法在积分环节中对从时间0到当前时间的偏差进行积分，采样时间的输出与过去的每个时间相关，而控制器的输出对应执行器的实际位置，整个过程耗费内存空间，计算时间长[3-4]。……