Deep Multi-Object Tracking

Multiple Object Tracking Using Deep Learning and Kalman Filter.

The tracker implemented combine the following two papers:

1. https://arxiv.org/pdf/1701.01909.pdf
2. https://arxiv.org/abs/1602.00763

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.

Overview of Files

Deep Apperance Model

• 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 Tracking

• kalman_filter_tracker.py: Taken from the implementation of the original author.

Deep Tracking

• deep_tracker.py: A new class which novelly combine both trackers to consider both apperance and motion featuers.