Repetitive face detection

Part of the job for identifying the suspect is to watch hours records for times. In this movie, they also watched the records in the lobby to find suspects with abnormal activities and the person who mailed.

To translate it into a computational problem, the input is a long video, the output is a subset of the video with suspect, the problem is to identify the subset with very low false-negative and could tolerate high false positive. The prior information we can make use of is that the face that was in the movie frequently in the continuous-time slots.

Pipeline:

• get fact detection

• For each image, get a list of the detected face

• get face similarity score

• find cluster