At present, disinfection is an important means to implement the same prevention measures of "people, objects and environment". Accurate and standardized disinfection work is related to the implementation of the overall measures of epidemic prevention. To this end, the project based on ROS and deep learning technology to design an intelligent disinfection vehicle, comprehensive consideration of epidemic prevention needs, to achieve fixed-point cruise killing, pedestrian detection, mask recognition and remote central control interface visualization and other functions. For the cruise killing function, this project takes Jeston Nano as the core control panel and completes the map construction of indoor scene based on the Gmapping algorithm of ROS system. Then, by using the odometer, tf coordinate transformation tree and radar point cloud and other data, based on the Navigation function package to achieve multi-point navigation of the car. For pedestrian detection, the project adopts the current mainstream YOLO v5-s detection algorithm, and completes model training in the pedestrian data set to achieve efficient detection of pedestrian targets, and combines with depth cameras to achieve accurate social safety distance measurement. For the mask recognition algorithm, the project uses the multi-task convolutional neural network MTCNN to achieve face detection and five-point calibration, and combines the lightweight convolutional neural network ShuffleNet v2 to achieve fast and accurate classification of masks. For the remote central control interface, the project is based on PyQt for interface development. The interface includes two display areas, which respectively display the real-time picture of the depth camera and the SLAM map. Users can navigate through the remote interface.