Background Seizure detection in epilepsy patients may facilitate diagnosis and improve diagnostic accuracy, thereby improving medication adjustment. Video-based systems can be a useful alternative to other modalities in the state of the art by providing live video feeds for caregivers and more information to health professionals, but they may raise questions related to patient privacy. Therefore, the main objective of this work is to create an accurate, affordable, and unobtrusive video-based seizure detection device that preserves patient privacy by design, which we called LAMPSY. Materials & Methods LAMPSY consists of a small processing unit integrated within a light fixture, with infrared and visible lighting modules and a small camera. A novel video-based seizure detection algorithm based on Optical Flow, Principal Component Analysis, seizure movement isolation with Independent Component Analysis, novel seizure metrics, and machine learning classification was developed. This method was tested on a set of 21 tonic-clonic seizure videos from 12 patients using Leave-One-Subject-Out Cross-Validation. Results & Conclusions State-of-the-art accuracy levels were achieved, with sensitivity and specificity of 99.06% +/- 1.65% at the equal error rate, and an average latency of 37.45 +/- 1.31 seconds. Seizures were segmented with an error of 9.69 +/- 0.97 seconds when compared to annotations by health professionals. The developed method preserved patient privacy by design due to its generation of unidentifiable and non-invertible representations of the data using Optical Flow, enabling it to function in real-time without saving video files. Moreover, it was robust to different lighting conditions, partial occlusions of the patient, and other movements in the video frame, thereby setting the base for accurate and unobtrusive seizure detection. Further work is currently being carried out to test the accuracy of this device in more patient populations.