Browse Prior Art Database

DEEP TRACK: A VIDEO CONTENT ANALYSIS SYSTEM

IP.com Disclosure Number: IPCOM000250598D
Publication Date: 2017-Aug-07
Document File: 6 page(s) / 387K

Publishing Venue

The IP.com Prior Art Database

Abstract

A technology for deep track assisted video content analysis system is disclosed. The system continuously monitors the patients and alerts the hospital staff in case of immediate medical attention. The system uses the provided video data and helps to track patient behavior through body movements and facial expressions. Therefore, on the basis of observations combined with deep learning the system identifies an unusual or emergency state. For example, the system can identify if a patient is going through a stroke or an epileptic fit and hence sends an alert to the required personnel.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 54% of the total text.

DEEP TRACK: A VIDEO CONTENT ANALYSIS SYSTEM

BACKGROUND

 

The present disclosure relates generally to video content analysis system and more particularly to deep track assisted video content analysis system.

Video content analytics (VCA) is the capability of automatically analyzing video to detect and determine temporal and spatial events.

However, there is a requirement for developing VCA algorithm that can be integrated with present Tele-ICU (intensive care unit).

It would be desirable to have an efficient technology for development of VCA algorithm.

BRIEF DESCRIPTION OF DRAWINGS

Figure 1 depicts a line diagram for deep track assisted face detection and motion detection.

Figure 2 depicts the detection of face by deep track.

Figure 3 depicts the emotion recognition by deep track.

Figure 4 depicts the motion detection by deep track.

Figure 5 depicts the motion tracking by deep track.

Figure 6 depicts the performance analysis for deep track.

Figure 7 depicts the difference between openCV (open source computer vision library) and dlib.

Figure 8 depicts the deep track capability to comprehensively describe the whole scene from a given frame of video.

DETAILED DESCRIPTION

A technology for deep track assisted video content analysis system is disclosed.  The system continuously monitors the patients and alerts the hospital staff in case of immediate medical attention. The system uses the provided video data and helps to track patient behavior through body movements and facial expressions. Therefore, on the basis of observations combined with deep learning the system identifies an unusual or emergency state. For example, the system can identify if a patient is going through a stroke or an epileptic fit and hence sends an alert to the required personnel.

As depicted in figure 1, deep track enables to capture the face given in a frame of video using histogram of oriented gradients (HoG) and support vector machine (SVM). The deep track employs dense optical flow algorithm to track persons in a subsequent video frames. Hence, deep track does not recognize faces rather preserve anonymity. Therefore, deep track identifies emotions of the actors in the frame from their expressions and detects motions present in the video given in a time frame. Also, deep track describes in natural language a given video scene.

Figure 1

Figure 2 depicts the detection of face by deep track. Dlib is an open source library in C++ and python. Dlib employs a sliding window based approach hovering over the frame. The system extracts HoG features and finally classify each boinding box into face or no face region. Therefore, HoG features are sensitive to colors, face related features like eyes, nose and their contours.

Figure 2

As depicted in figure 3, the emotion recognition by deep...