Page 222 - AI for Good-Innovate for Impact Final Report 2024

AI for Good-Innovate for Impact

Use case – 52: Computer Network Fusion Video Brain

Country: China

Organization: China Mobile Communications Corporation Co., Ltd

Contact person: Zhanmei, Zhang; 13802881237@139.com

52�1� Use case Summary Table

Domain Industry, Innovation, and Infrastructure
The Problem to be addressed • Massive video data is mainly monitored through manual
viewing, which requires a lot of human resources and
cannot monitor video content in real time.
• The analysis of massive video data in the central node
consumes a lot of network bandwidth resources.
• For intelligent video analysis algorithm training, the lack
of sample data makes it difficult to improve the accuracy.
• Video analysis only through the traditional target detec-
tion small model is prone to produce a large number of
false positives, requiring a lot of manpower to audit.

Key aspects of the solution • In order to meet the business expansion needs of
massive video access, large models combined with small
models will be introduced to complete video analysis:
1. Based on the combination of high reference quantity
and strong feature capture ability of large model and
high flexibility of small model, it can effectively realize
efficient analysis of video.
2. In order to use large models as feature extractors,
perform preliminary analysis of videos, infer image
events and behaviors, and extract useful feature informa-
tion from videos, such as color, texture, shape, motion,
etc. The feature information extracted from the large
model is further analyzed and predicted by using the
small target detection model

Technology keywords Cloud edge collaboration; Large and small model collabo-
ration

Data availability https:// huggingface .co/ datasets; https:// public .roboflow
.com/

Metadata (type of data) Structured and unstructured data

206