Deep learning for object identification in ROS-based mobile robots

In this paper, an open source robotic middleware ROS is applied to identify objects with a Raspberry Pi based mobile robot. Faster R-CNN algorithm is considered for enhanced deep learning. The capabilities of robot control and object detection are verified through experimental validations. Based on a low-cost Raspberry Pi control kernel, an easy-to-use interface is developed to integrate front-end and back-end applications. Firstly some ROS packages for the mobile robot are needed to be designed. Then the captured environment information will be provided for objective detection using GPU-accelerated computing. In this paper, Kinect sensor is used to capture images and the deep learning algorithm is implemented based on ROS middleware. All previous mentioned function modules are contained in a cloud service.