基于CNN的车载终端自动化测试系统设计与实现

Design and Implementation of Automated Test of Vehicular Terminal Based on Convolutional Neural Network

随着人工智能技术飞速发展,机器学习对于提升自动化测试效率,改善电子产品质量起着至关重要的作用。针对传统车载终端(车机)测试方法效率低下,人工成本高的问题,提出了一种利用ResNet50_Sub_Pixel_DSNT(RSPD)+透视转换矫正+YOLOv4的测试方法。该方法利用RSPD实现对车机屏幕4个关键角点的定位,并通过透视转换算法对定位目标区域进行截取矫正,采用YOLOv4模型对矫正后的图片进行相关功能控件位置识别,驱动机器臂点击识别区域目标。测试结果表明,RSPD+透视变换矫正+YOLOv4的测试方法,能够准确控制机器臂完成车机屏幕功能控件的点击测试操作,其准确度达到97.3%。

With the rapid development of AI technology,machine-learning plays a vital role in improving the efficiency of automated test and the quality of electronic products.A test method using ResNet50_Sub_Pixel_DSNT(RSPD)+perspective conversion correction+YOLOv4 is proposed to solve the problems of low efficiency and high labor cost in traditional test method of vehicular terminal.RSPD is used to locate the four key angular points of screen,then the target area is captured and corrected based on the perspective conversion algorithm.Finally,YOLOv4 model is used to locate the associated functional controls based on the corrected picture,driving the robot-arm to click on the target controls.The results show that the accuracy is up to 97.3%.The method based on RSPD+perspective conversion correction+YOLOv4 can accurately instruct the robot-arm to perform the test operation by clicking on the functional controls on the vehicle screen.