Indoor-Tracker (in development...)

Intro:

What the Indoor-Tracker suppose to do:

The Indoor-Tracker will be a indoor-path-recording-application implemented with python. 

The input will be several camera video streams located around a conference hall.

During registration, a 128-dim face feature map will be collected using dlib-face-recognition from each visitor.

A YOLO-based body tracking will be applied to the video streams to take down the trace of each person detected. 

Each trace without identification will be stored as temporary data. 

Whenever the camera got a clear shot of a face of a visitor, the face data will be analyzed and 

matched with the dataset collected during the registration to identify the owner of that trace. 

Once identified, the trace will be added to the record of that specific visitor.

Something important to know:

The Indoor-Tracker is still under development, and it heavily depends on other opensource frameworks and projects.

Currently it is based on deep_sort_yolov3, which can be found at: 
    https://github.com/Qidian213/deep_sort_yolov3
    
For dlib-face-recognition, download the model from:
    http://dlib.net/files/shape_predictor_5_face_landmarks.dat.bz2\n
    http://dlib.net/files/dlib_face_recognition_resnet_model_v1.dat.bz2
    
Only modified and added files are provided due to copyright issues.

Instructions:

1. Download the deep_sort_yolov3

2. Replace origin files with files provided

3. Download face-recignition models, modify the path in demo.py

4. Test it with a video (Only for now, just one video stream)

Environments:

OS: Ubuntu 16.04

Python: 2.6+

Dependency: dlib, keras