ACCIO

System Architecture

Getting started

install docker-compose

first time setup

docker-compose up -d
scripts/init.sh
pip install -r requirements.txt

for database migration and setup:

flask db migrate
flask db upgrade

to login to accio db inside the container:

docker-compose exec db bash
psql accio -U postgres

run the server

flask run

To run yolo object detection you have to donwload the yolov3.weights Then add yolov3.weights in

plugins/yolo_object_detection/yolo_coco

Implementation Details

Tables

• Videos
  • id
  • name
  • results
  • path/url
  • processed
• Plugins
  • id
  • name
  • is_enabled
  • executable_path
  • system_configuration
  • plugin_configuration

Services, Classes and Directory Structure

• api/ - flask api serving the frontend

• engine/ - module carrying all the logic happening behind the scenes

  • datafeed::DataFeeder
  • keyframes::KeyFrameExtractor
  • plugins::PluginManager
    • abstracts the interaction with plugins
  • engine::EngineManager
    • discover video from datafeed
    • for each plugin in plugins list
      • perform keyframes extraction on video (depending on its system configuration)
      • form input obj (either output of keyframes or raw video)
      • output = PluginManager.run(plugin, input_obj)
      • Indexer.index(video, output)
  • indexer::Indexer
    • reflect output results to video db records
      • output: (view only)

        {
          "src_name": "xxx",
          "results": "xxx"
          [
            {
              "from": "xxx",
              "to": "xxx",
              "content": "xxx",
              "bb": [],
              "confidence": "xx" 
            }
          ]
        }

• plugins/ - repository of all implemented plugins

  • each plugin inherits from an AbstractPlugin class, implementing a run method and configurationDict field

Screens

Client UI

A search box that sends the query to /api/search?q="xxx"x and renders the results with potential interaction with the videos and their frames

Admin UI

A place to manage plugins and their configuration

subsystems

plugins integrator

import paths:

• plugins.facedetection.facedetection.FaceDetectionPlugin
• plugins.yolo_object_detection.yolo.YoloPlugin

run

python engine/plugins_integrator.py --name <plugin name> --executable_path <import path>

python engine/plugins_integrator.py --name face_recognition --executable_path plugins.facedetection.facedetection.FaceDetectionPlugin
python engine/plugins_integrator.py --name yolo --executable_path plugins.yolo_object_detection.yolo.YoloPlugin               

data feeder example

run

python engine/datafeeder.py --sleeptime <time in secs> --repository <relative path>

python engine/datafeeder.py --sleeptime 20 --repository ./repository

Peresentation

• Talk about input type (scene, raw video, keyframes [x])

objects

Chair, Car, Person, Jim, Stanley