The Applications of Image and Video Quality Assessment for Visual Surveillance, MOST, Taiwan. 2013/8-2016/7.
- This is a three-year project to research the application of image and video quality assessment for visual surveillance. The research topic includes three research topics: single image de-fog, camera anomaly detection, and moving object detection for night surveillance. Therefore, the goal of this project is not only to study the three research topics, but also apply the image and video quality assessment to the three topics...
Vision-based Sleep Disorder Analysis. MOST, Taiwan. 2010/8-2012/7.
- Sleep disorder has been a significant symptom in sleep medicine. The diagnosis of sleep disorder is obtained by analyzing the recorded biosignal data from PSG, polysomnography. The PSG records all important biosignal and video data of patients when they sleep over night. The PSG is actually an integration of medical instruments more than 10-channel biosignals. It provides biological details of the patient for the diagnosis of sleep disorder. However, the PSG is very complicated and hard to operate. Usually a well-trained physician is necessary to help the wearing of PSG electrodes onto the patient's body. Therefore, the measurement and diagnosis by PSG cost high medical resources.
- The goal of this sub-project is to study a new measurement method for the sleep disorder analysis. The study will focus on video signal, and use vision-based approach for the automatic analysis. The video of sleep will be infrared images of night vision. The infrared images are analyzed by computer vision techniques to obtain the body motion and gesture of the patient. The data of video analysis can replace many biosignals, and then reduce the number of recording biosignals. Also, the high capacity requirement of video data due to 8-hour over night recording can be alleviated by vision analysis. Video summarization and key frame extraction will be applied to the over-night video to extract important video clips for efficient storage, browsing, and retrieval.
- The three-year study of this sub-project includes three research topics. In the first year of this sub-project we will develop a sleep/wake classification method by infrared image analysis. Methods of Noise filtering, image enhancement, and human detection will be studied. The second year will be devoted to examining pose recognition of night sleep. Object detection, human body segmentation, and pose estimation will be studied. Integration of multimodal signals will be researched in the third year. The signals of infrared sensor, actigraph, and oximetry will be combined for the analysis of important sleep status. Key frames and clips will be extracted as video summarization of an over-night sleep video.