基于无人驾驶小麦收割机立体视觉感知系统

Stereo visual perception system based on unmanned wheat harvester

针对小麦收割机在农场无人驾驶作业时无法实现动态障碍的实时避障,无人驾驶技术安全性低等问题,设计一种基于立体视觉与深度学习相结合的无人驾驶立体视觉感知系统。首先使用立体视觉相机采集左右目灰度图像,通过图像中像素位置的视差以及立体视觉成像原理,实现对障碍物的距离计算;再将相机采集的RGB图像通过深度学习进行处理,实现障碍物的检测识别,最终完成对动态障碍物的感知。结果表明,基于立体视觉与深度学习的无人驾驶感知系统在农场无人驾驶作业中动态障碍物的检测速率达到30.1 fps,精确率达到98.24%。该方法能够较好的满足作业中动态障碍物检测的识别要求,显著提升无人驾驶小麦收割机作业时的安全性和可靠性,为智能农机无人驾驶的研制奠定理论与技术基础。

In response to the problems of wheat harvesters being unable to achieve real‑time obstacle avoidance of dynamic obstacles during unmanned operation on farms,and the low safety of unmanned driving technology,this paper designs an unmanned stereo vision perception system based on a combination of stereo vision and deep learning.The system first uses a stereo vision camera to collect grayscale images of left and right eyes,and calculates the distance between obstacles through the disparity of pixel positions in the image and the principle of stereo vision imaging;Then,the RGB images collected by the camera are processed through deep learning to achieve obstacle detection and recognition,ultimately completing the perception of dynamic obstacles.The research results indicate that the autonomous driving perception system based on stereo vision and deep learning has a detection rate of 30.1 fps and an accuracy rate of 98.24%for dynamic obstacles in unmanned driving operations on farms.The method proposed in this article can effectively meet the recognition requirements of dynamic obstacle detection during operation,significantly improving the safety and reliability of unmanned wheat harvesters during operation,and laying a theoretical and technical foundation for the development of intelligent unmanned agricultural machinery.