The technology of 3D recognition is evolving rapidly, enabling novel applications towards human-centric intelligent environments. On top of these applications, tracking human sitting posture is essential for realizing human comfort and healthy environments. However, existing techniques rely on cameras or chair-attached sensors, which are privacy-invading or non-common technology in every environment. This paper introduces a ubiquitous portable technology for tracking sitting posture with a plug-and-play concept. Specifically, at the core of the proposed system, we leverage our proprietary LiDAR device to scan the human’s sitting posture and render it in a privacy-keeping point cloud representation.The capture point cloud samples are leveraged to train an efficient deep neural network for enabling accurate recognition of the sitting posture. The proposed network significantly reduces the computational complexity of the model by learning special features that simplify the classification task. We implemented and evaluated the proposed system on nine different human postures in a real-world environment. The results show that it obtains an accuracy of 87% with a drastically reduced processing time.