Get Started with Video Labeler 

Introduction

Object recognition and pursuit are among the foremost everyday and sophisticated tasks that robotic sensing systems should perform to support complex psychological feature tasks that love distinguishing between events and actions. The goal of seeing is to establish the object's gift within the 3D scene accurately. This is a difficult task as the scene could be cluttered, objects obscuring each other, and there could also be lighting or different angles. Object and position recognition is essential in several sensible applications such as industrial plant automation, navigation, vehicles, elements inspection, CAD / CAM, etc. Our primary interest is in object recognition for autonomous vehicles applications in a road environment in this report. We are monitoring people, robots, cars, and pedestrians. We want to observe people and these objects to improve the performance and safety of vehicles, so vehicles and similar systems can interact with their environment. We emphasize Position, movement, pose, and classification data to determine if a person is crossing the street or a specific car is on the move.

For this application, we will be using the inbuilt Video Labeler App in MATLAB. This app facilitates us by creating labels around the region of interest. And the data produced by this app can be used for training algorithms such as object detectors, image classifiers, etc.

Importing Data

Labelling

 After importing, start by navigating to the required frames or select the portion of the video. Now define the type of label you intend to draw. It can be a rectangle, polygon, line, pixel, or projected cuboid. Click on Add labels, enter a name, select the type, color, group, and description. After that, you can choose a region of interest by clicking and dragging over the region. You can also create sub-labels, attributes, and scene labels.

Sublabels

It is a label that belongs to the parent label. They are used for more detailing and can be created by clicking on the sublabel. Follow the same steps as while adding a label. For example, a man is a label whose features like eyes and nose can be sub-labels.

Attributes

It specifies additional information about labels or sub-labels. For example- A man's height can be an attribute. It can be defined in following types:

Numeric Value — Specify a numeric scalar attribute

String — Specify a string scalar attribute

Logical — Specify a logical true or false attribute

List — Specify a drop-down list attribute of predefined strings

To add an attribute, click on Add Attribute and fill in the information.

Scene Label

This option greatly benefits adding environment/scene additional information such as events in the background. Click on Scene Label, then provide name, group description, and colour optionally. It can be added to a frame or to a time interval.

Video Labeler

As of now, we have labelled a single frame of our video. Now we can proceed in two ways, either manually or using an algorithm for automation:

Labelling Manually

This is a challenging but efficient method for labelling. In this, you must create labels, sub-labels, attributes in each frame individually. It will not carry this information from one frame to another.

Labelling with Automation Algorithms

After completing your work, you can use the View Label Summary option to see label frequency, scene labels, and many other things.

Exporting Data

Conclusion

We have completed importing data, acting on it, and at last, exporting it to Workspace. in a while, we have trained the model and tested it, which ends in 1.0 precision. It's prompt to use labels with correct form and sizes; otherwise, it'll lack accuracy.