gemblerz / plugin-trafficstate

YOLOv4

• Functions for YOLOv4 are in the foler tool.
  
• The functions in yolov4.py are to detect vehicles.
  
• The class information is in detection folder -- the coco classes.

DeepSort

• Functions for DeepSort that tracks vehicles are in siamease_net.py (main algorithm to track vehicles based on their features) and the folder deep_sort.
  
• The functions are called at deepsort.py which is main class to run vehicle tracking.
  
• Traffic state calculation are performed through functions in deepsort_plugin.py.
  
• Traffic state calculation and vehicle tracking are based on the object detection results from YOLOv4, so that the deepsort_plugin.py cannot be excuted by itself -- they requires input values that came out from yolov4.

PyWaggle

• We use the two methods (YOLOv4 and DeepSort) to calculate traffic state.
  
• And the calculation results are sent through PyWaggle to Cloud (for detail information see PyWaggle).
  
• The PyWaggle is simply implemented in app.py, and still testing the modules (3/27/2021).
• To build the plugin as a Docker container to run in Waggle nodes, refer exmple plugin and a wiki (3/31/2021).

Notes to developer:

• Timestamp for each traffic state needs to be the time when the video captured: The timestamp must be provided with the video.
• The waggle.plugin.publish function is tested, and checked with log.
• Not yet plugin-ized (3/30/2021): needs to be dockerized and create sage.json and others using virtual waggle