Video Analytics Engine
based on Object Labelling

Supervisor: Dr. Chung Ronald
Student: TSE Wai Shun Vincent

Learn More

Introduction

Artificial Intelligence is a significant target and will be a milestone of the computer science field, while computer vision and deep learning are the basic requirements in proceeding to the final working stage of artificial intelligence. Object labelling is a kind of computer vision that making the computer know how to ‘see’, that makes computer could recognize different objects that appeared in an image so as a video. Other than the traditional pixel-based object labelling method that segment the image into superpixels, this project will realize the video engine by the aid of category tag based object labelling that splits a single image into different regions and predicts the boundary of potential objects for each region(Redmon, 2017), which is very fast compare to traditional approach in the past that the object detection speed is currently up to 220 frames per second(FPS). As only 25 FPS is required for a smooth animation or movie, that means such object labelling technology is applicable be used in real time object detection in a playing video.

Mid-October

Iteration 1.1
Structural frame of the desktop application (GUI, able to perform input and output)

Mid-November

Iteration 1.2
Apply the image object labelling to the whole inputted video and output the detection result.

Late-December

Iteration 2.1
Application is able to extract useful information from the detection result of the video.

Mid-February

Iteration 2.2
Application Completed

Late-March

Testing of application completed

16-20 April 2018

Final presentation 

--> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> --> -->