TABLE OF CONTENTS

Overview

Today, Artificial Intelligence has revolutionized the way people have been looking at information collection and distribution. Detecting objects within videos for enhanced searchability within the video by navigating to time stamps where object appeared and to redact all those objects without manual human intervention and labor. 


Before you start

  • Make sure you log into your VIDIZMO Portal as an Administrator or Manager to be able to configure VIDIZMO Indexer in VIDIZMO Portal for object detection and enhanced searchability


VIDIZMO Portal Configuration

1. From the Portal's Homepage:

i. Click on the menu icon on the top left-hand corner of the screen to bring up the left navigation pane

ii. Then click on the down arrow to expand the Admin section

iii. Select Portal Settings to open the Portal Settings navigation panel


2. From the Portal Settings navigation pane:

  1. Click on the Apps to expand the list of application services that are available in VIDIZMO
  2. Select Content Processing where you can set up VIDIZMO Indexer
  3. Click on the settings icon against VIDIZMO Indexer to configure VIDIZMO Indexer and select the objects



3. From the VIDIZMO Indexer screen: 


The configuration screen of the VIDIZMO Indexer is designed in a way that portrays the dependency of every field on the selected object detection category, which are as follows:


a. Detection Class

This is the type of object that can be detected by VIDIZMO Indexer 


Following fields need to be filled for every detection class to configure VIDIZMO Indexer

  1. Model Size: Three different models to choose from i.e. "Small", "Medium", and "Large".
  2. Confidence Threshold: Input a value between 25-95.
  3. Tracking Frame: Input a value between 7-25. 
  4. Category: Choose from the category (if applicable)


Note: Model size, Confidence Threshold, and Tracking Frame fields are explained below. 


4. Click on Save Changes once you are done with configuration.

5. From the Content Processing screen:

  1. Enable the toggle button to configure automatic object detection video in your portal.


A notification will appear briefly stating Portal Settings Updated Successfully.  


What is Model Size, Confidence Threshold, and Tracking Frames


Model Size

There are 3 models i.e. Small, Medium, and Large.

Small Model: This model is less resource hungry, which means it requires less memory and computation power and it takes lesser time to detect objects within the video, however, this model detects objects with less accuracy as compared to the medium and large models. This model is recommended for users who have machines with less computation power and memory. 


Medium Model: This model is more resource hungry as compared to small model, which means it requires more memory and computation power and it takes more time to detect objects within the video as compared to small model, moreover, this model detects objects with more accuracy as compared to the small model. This model is recommended for users who have machines with moderate computation power and memory. 


Large Model: This model is most resource hungry, which means it requires more memory and computation power and it takes more time to detect objects within the video as compared to small and medium models, moreover, this model detects objects with most accuracy as compared to the small and medium models. This model is recommended for users who have high-end machines (GPU is a plus). 


Confidence Threshold

Confidence Threshold is the field that the user can set to tell the VIDIZMO Indexer to only show the detected objects when the model is confident at a minimum or more than the provided input value. The model will only save those object on which it is at a minimum or more confident than the provided input value in this field and disregard otherwise.


Tracking Frames

Tracking is the field that the user can set to specify the model to consider it as an object when the specific object appears in for e.g. at least 7 consecutive frames. The model will only consider those as an object which appear in atleast 7 consecutive frames and disregard those objects that don't appear in 7 consecutive frames.