The watch world identifies anomalies or unexpected occurences in video data. It does this by building the data into a hierarchy of features and learning patterns of feature activity.
To run BECCA with the watch world, clone this repository into your local copy of the becca repository so that it sits in the same directory as the core and worlds directories.
In tester.py, add the line:
from becca_world_watch.watch import World
and comment out all other World import lines. The watch world also makes use of the OpenCV libraries through their python bindings, it imports cv2.
Typing python tester.py at the command line will run the listen world. It draws training data from all .flv and .mp4 files that it finds in the becca_world_watch/data directory.
In watch.py, manually set the flag variable self.TEST to True when you want to test the anomaly detection performance. It will look for the test data in becca_world_watch/test/test_long.avi and for ground truth information in the same directory under truth_long.txt.
The ground truth text file format is two ASCII numbers per line (e.g. 12.3 14.7) indicating the start and stop times of an anomaly in seconds. There is one line per anomaly.
Adjust self.fov_horz_span and self.fov_vert_span in watch.py to change the number of sensor columns and rows, respectively.