A smile/laughter recognition mechanism for smile-based life logging

Most situations that cause people to smile are important and treasured events that happen in front of the other people. In life-logging systems that record everything with wearable cameras and microphones, it is difficult to extract the important events from a large amount of recordings. In this research, we design and implement a smile-based life-logging system that focuses on smile/laughter for indexing the interesting/enjoyable events on a recorded video. Our system, features an original smile/laughter recognition device using photo interrupters that is comfortable enough for daily use and proposed an algorithm that detects smile/laughter separately by threshold-based clustering. The main challenge is that, since the reasons people smile and laugh are quite diverse, the system has to detect a smile/laughter as different events. Evaluation results showed that our mechanism achieved a 73%/94% accuracy in detecting smile/laughter, while actual use of the system showed that it can accurately detect interesting scenes from a recorded life log.