Lightweight intelligent autonomous unmanned vehicle based on deep neural network in ROS system.

Due to the huge amount of data and calculation consumption produced by the automatic driving system, the performance of the hardware platform is required to be higher. Therefore, this paper proposes a lightweight autonomous vehicle scheme based on deep neural network and ROS system. Open CV image processing technology is used to collect simulated lane data by cameras to extract its features, and traffic signal signs and other data are taken. The Tensor Flow Lite neural network framework is used to build the Faster R-CNN target detection algorithm, train the convolutional neural network, and enable it to recognize traffic signals and make corresponding decisions automatically. The experimental results show that the unmanned vehicle has the ability of lane recognition and feature extraction, and has the ability of traffic signal recognition and automatic driving to a certain extent.