Multiple Object Tracking Using Deep Learning and Kalman Filter.
The tracker implemented combine the following two papers:
The first paper present a state of art deep learning model to do multiple object tracking. The second paper present a simple online Kalman Filter tracker.
The model implemented in this repo novelly combine both tracker, which can do online multiple object tracking using a state of art deep learning model to identify the apperance features and a simple online Kalman filter tracker to identify motion features.
data_input.py
: Script to preprocess the MOT15 data.train.py
: Contains training process of the apperance network.apperance_network.py
: Contains the deep apperance model architecture.
kalman_filter_tracker.py
: Taken from the implementation of the original author.
deep_tracker.py
: A new class which novelly combine both trackers to consider both apperance and motion featuers.