基于双目视觉的泊车机器人障碍物检测系统
|
摘要
针对泊车机器人和智能停车库研究领域中对视觉系统的需求,设计了一种基于双目视觉的泊车机器人障碍物检测系统。借鉴物理学中的控制变量法完成双目相机标定,采用Bouguet算法进行立体校正,引入YOLO卷积神经网络对障碍物进行快速检测,利用改进立体匹配算法完成对弱光照下光滑边缘障碍物的检测,搭建双目视觉系统并进行实验验证。实验结果表明,该系统检测平均耗时为0.463 s,在1 400 mm至2 100 mm范围内检测误差在50 mm内,具有良好的实时性和较高精度,为经济型泊车机器人的研制奠定基础。
Aiming at the demand of vision system in the research field of parking robot and intelligent parking garage,a parking robot obstacle detection system based on binocular vision is designed. The binocular camera calibration is done by using the control variable method in physics. The Bouguet algorithm is used for stereo correction. The YOLO Convolutional Neural Network is introduced to detect the obstacles quickly. The improved stereo matching algorithm is used to detect the smooth edge obstacles under weak illumination,set up a binocular vision system and conduct experimental verification. The experimental results showthat the consumed detection time of the system is 0.463 s on average,and the detection error is within 50 mm from 1 400 mm to 2 100 mm.It has good real-time performance and high precision,which lays a foundation for the development of economical parking robot.
Aiming at the demand of vision system in the research field of parking robot and intelligent parking garage,a parking robot obstacle detection system based on binocular vision is designed. The binocular camera calibration is done by using the control variable method in physics. The Bouguet algorithm is used for stereo correction. The YOLO Convolutional Neural Network is introduced to detect the obstacles quickly. The improved stereo matching algorithm is used to detect the smooth edge obstacles under weak illumination,set up a binocular vision system and conduct experimental verification. The experimental results showthat the consumed detection time of the system is 0.463 s on average,and the detection error is within 50 mm from 1 400 mm to 2 100 mm.It has good real-time performance and high precision,which lays a foundation for the development of economical parking robot.
引文
[1]徐欣,周香琴,江先志,等.基于物联网技术的小区停车位共享平台的设计与开发[J].工业控制计算机,2018,31(1):139-141.
[2]申爱萍.揭开"最牛泊车机器人"的神秘面纱[J].驾驶园,2017(9):42-43.
[3]刘爽.基于二维码识别的自动泊车机器人定位导航技术研究[D].武汉:华中科技大学,2017.
[4]研华科技.怡丰机器人携手研华:以技术力量,扩展AGV市场应用[J].自动化博览,2018,35(8):76-77.
[5]魏言华.基于视觉的车辆后方障碍物检测算法研究与实现[D].沈阳:东北大学,2008.
[6]谢若冰.双目立体成像和显示的FPGA视频处理技术研究[D].北京:北京理工大学,2015.
[7]叶峰,王敏,陈剑东,等.共面点的摄像机非线性畸变校正[J].光学精密工程,2015,23(10):2962-2970.
[8] Shubham Shinde,Ashwin Kothari,Vikram Gupta. YOLO based Human Action Recognition and Localization[J]. Procedia Computer Science,2018,20(3):133-134.
[9]王昊.基于卷积神经网络的目标检测与识别方法研究[D].南京:南京财经大学,2016.
[10]施泽浩,赵启军.基于全卷积网络的目标检测算法[J].计算机技术与发展,2018,28(5):55-58.
[11]杨梓豪.基于区域卷积神经网络的物体识别算法研究[D].北京:北京邮电大学,2017.
[12]李旭冬.基于卷积神经网络的目标检测若干问题研究[D].成都:电子科技大学,2017.
[2]申爱萍.揭开"最牛泊车机器人"的神秘面纱[J].驾驶园,2017(9):42-43.
[3]刘爽.基于二维码识别的自动泊车机器人定位导航技术研究[D].武汉:华中科技大学,2017.
[4]研华科技.怡丰机器人携手研华:以技术力量,扩展AGV市场应用[J].自动化博览,2018,35(8):76-77.
[5]魏言华.基于视觉的车辆后方障碍物检测算法研究与实现[D].沈阳:东北大学,2008.
[6]谢若冰.双目立体成像和显示的FPGA视频处理技术研究[D].北京:北京理工大学,2015.
[7]叶峰,王敏,陈剑东,等.共面点的摄像机非线性畸变校正[J].光学精密工程,2015,23(10):2962-2970.
[8] Shubham Shinde,Ashwin Kothari,Vikram Gupta. YOLO based Human Action Recognition and Localization[J]. Procedia Computer Science,2018,20(3):133-134.
[9]王昊.基于卷积神经网络的目标检测与识别方法研究[D].南京:南京财经大学,2016.
[10]施泽浩,赵启军.基于全卷积网络的目标检测算法[J].计算机技术与发展,2018,28(5):55-58.
[11]杨梓豪.基于区域卷积神经网络的物体识别算法研究[D].北京:北京邮电大学,2017.
[12]李旭冬.基于卷积神经网络的目标检测若干问题研究[D].成都:电子科技大学,2017.