Page 346 - Big data - Concept and application for telecommunications

6 Big data - Concept and application for telecommunications

6 Application scenarios

6.1 Intelligent video retrieval
It is difficult for traditional video surveillance to retrieve a specific object in massive video data. However, a
user can use text, picture or video as input to search in a large number of videos according to the various
characteristics of the objects in visual surveillance enhanced by big data (VSBD). Two examples follow.
a. Using the image of the target person as input, users can quickly search for similar objects in massive
videos according to the facial, clothing colour distribution or body features of the image.

b. Using characteristic information such as licence plate, vehicle type, colour, brand type or body
feature of the target vehicle as input, users can retrieve similar objects in multiple videos with the
trajectories displayed.

6.2 Intelligent event detection

VSBD can automatically identify specific events (fire, fake licence plate, etc.) that users are interested in based
on various data. The data can be derived from multiple video sources or video and sensor data combinations.
The criteria for identifying specific events can be predefined rules or models learned from machine learning
(ML).
For example, VSBD can detect a fire disaster by analysing the related real-time videos as well as sensing data
from smoke detectors and temperature sensors in areas of concern to users.

6.3 Status prediction

The value of traditional video surveillance data has not been explored. VSBD can effectively exploit the value
of historical data. It can extract a large amount of effective information from video surveillance data and be
used in many aspects of prediction. For example, according to the data collected by cameras, VSBD can
analyse the rules and forecast future traffic flow in order to achieve efficient traffic control.

7 Service requirements

7.1 Video retrieval

7.1.1 Object retrieval
SRV001: VSBD is required to support various conditional retrieval inputs and search objects in massive video
data across cameras by different abstract features rapidly.
SRV002: VSBD is required to support continuous video capture for selected objects.

SRV003: VSBD is required to adapt automatically despite scenarios, illumination, camera angles and position.

7.1.2 Event retrieval
SRV004: VSBD is required to support retrieval of the characteristics for particular events, such as people
gathering in time and space. A VSDB is required to support the statistics for the selected range of alarm
events.
7.1.3 Integrated retrieval

SRV005: VSBD is recommended to support real-time and historical trajectories video retrieval from real-time
video and historical database by the conditions of time, space, hot events, features marks and object people
or vehicles.

7.2 Statistical analytics

SRV006: VSBD is required to support data collection for vehicle classification, trajectory tracking and
monitoring, and registration location.

338 Application