Research on autonomous driving perception based on deep learning algorithm

Multiple source sensor fusion is the foundation of motion planning for autonomous driving system, which is the crucial part in improving the performances for unmanned operational system. In this article, based on the deep learning platform CATARC constructed, applied with Udacity’s Lincoln MKZ multiple sensor data, implemented with Robotic Operation System, Computer Vision, PointCloud Library, Deep Neural Networks and Extended Kalman Filter, constructed a low-cost object pose estimation data fusion solution, aiming at technic support for the industrialization of autonomous driving technologies.