<p>In this work, we aim to achieve room-level localization for mobile robots in industrial workshops. It is difficult to obtain precise localization information via common methods because of the complexity of the industrial environment. Our findings show that precise room-level localization can be achieved via LiDAR-based point cloud registration and object recognition. For this purpose, we formulate room-level localization as a classification problem. Registration and object recognition are used to extract features from point clouds. After the data enhancement algorithm, called Stacked Auto Encoder is employed to overcome the issue of limited feature data, the neural network algorithm is leveraged to address the classification problem. To this end, we collected point cloud data from industrial workshops and performed experimental validation. We evaluated the recognition performance of the algorithm in a metallurgical workshop and achieved good accuracy.</p>