A Non-visual Sensor Triggered Life Logging System Using Canonical Correlation Analysis

Life logging is one of the key service in modern life as wearable devices are forming an emerging market. Life logging has advantages to enlarge the human memory and even can help patients who suffer from memory impairment. Periodical picture taking is the simplest and the most widely used method but inefficient in both energy and memory side. In this paper, we suggest a novel capturing points decision method using the combination of visual information and non-visual information. In order to merge visual information and non-visual information, we adopted canonical correlation analysis (CCA) which is a statistic technique to find the mapping function for two different domain data into a highly correlate domain. In this way, we showed the new possibility of life logging system with minimum heuristic methodology. Moreover, we tested our approach on restricted real life situation and evaluated the result based on an image diversity measure using a determinantal point process (DPP) and showed better results than conventional methods.