Multi Object Tracker with Visualization

Use Deep SORT algorithm to track objects in a video, then visualize the resulting tracks with simple streamlit web app.

Installation:

• Create python environment (python 3.9) and install the dependencies pip install -r requirements.txt
• Download YOLOv4 weights and put them in the model folder:
  • https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v3_optimal/yolov4.weights

Usage

• Run Object tracker python object_tracking.py on a video (default video is cars.mp4 in the data folder). New Database with tracking info should be created.
• To change the target video, put the video in the data folder and use the --video command line argument to specify the video path
• Run streamlit run visualize.py and open the web interface

Repo Overview:

• object_tracking.py: Track multiple objects in a video using deep sort algorithm. Specify video path with --video command line argument or leave it out for default data/cars.mp4 video.
• database.py: Features a simple interface to a sqlite database, used to write and read tracking information.
• mobilenet.py: Embed an image to a single dimensional vector, used for feature comparison in the deep sort algorithm.
• visualize.py: Visualize the resulting tracks in a simple streamlit web app. Select the video and trackID to visualize the whole trajectory or view a single frame and the matching bounding box, along with the detection confidence.
• data folder: Holds the videos and the class names for YOLO object detector.
• model_cfg folder: Holds the configuration files for YOLO models.
• model folder: Holds the weights for YOLO models.