Extract sequences of features of each person per video

[Asami-1]

Change the current lstsm_preprocess script to :

• Not do padding
• Extract each person and store it in json ./video_name/sequence/id, that is a data point for lstm.
• If more than half is a given class, we consider the whole sequence as the class

[create-issue-branch[bot]]

Branch issue-24-Extract_sequences_of_features_of_each_person_per_video created!