基于深度学习的汽车仪表标识辨别系统设计

The Design of Automotive Instrument Cluster Identification System Based on Deep Learning

本文利用硬件在环仿真和图像识别技术,开发了汽车仪表标识自动辨别系统。系统的硬件为d SPACE仿真控制平台,模拟CAN报文实现仪表标识显示并由摄像头完成图像采集,通过双变滤波算法滤除图像噪声,调节图片像素点的对比度和亮度,利用图像帧差法对标识进行定位提取,借助深度学习Inception网络对标识信息进行辨别,通过修改全连接层结构以适应标识的分类,并推导了网络中误差求解公式。为了便于应用,设计了用户交互界面,测试结果中系统的准确率达到了86%以上,起到了很好的分类和辨别效果,辨别结果转换为CAN报文反馈给仿真机柜,从而实现了汽车仪表功能的半自动化测试。

An automatic identification system is developed using hardware-in-the-loop simulation and image recognition techniques. The hardware platform is d SPACE simulation cabinet and the automotive instrument cluster displays information through simulated CAN messages. The image captured by a camera is processed by bilateral filtering algorithm and the contrast and brightness of the image are adjusted. Icons can be detected by employing the image sequences processing technology. T he icons are then identified by the deep learning network. Inception network architecture is adopted in depth learning and full connection layer parameters are modified to fit the identified classification. The formula for solving the network is derived. MFC user interface is designed for application. The accuracy rate of the system has reached over 86%,which has achieved good classification and identification. The result is converted into CAN messages and fed back to the simulation cabinet to realize semi-automatic test of automotive instrument clusters.