Help Center> > Service Overview >What Is Video Analysis Service (VAS)?

What Is Video Analysis Service (VAS)?

VAS uses AI technologies to perform intelligent analysis of video content. VAS includes a wide array of functions, such as video preprocessing, content moderation, content analysis, editing, search, and video fingerprinting. It delivers the detection, tracking, attribute identification, behavior identification, content moderation, video abstraction, and can tag the objects in videos. VAS efficiently and effectively analyzes your video content.

Currently, VAS provides the following subservices:
Table 1 VAS subservices




Video Content Recognition (VCR)

VCR detects and analyzes diversified objects in videos with all-inclusive algorithms. VCR's accurate segmentation of essential information greatly increases the effectiveness and efficiency of video content recognition.

  • Cost-effectiveness

    Ingests and analyzes video files and streams not only on the cloud, but also on edge nodes, which greatly reduces data ingestion costs.

  • High reliability

    Supports real-time and batch video analysis with high concurrency and low latency, covering complex scenarios such as extreme weather and dynamic camera angles.

  • High efficiency

    Supports customization and categorization of video qualities with the ability to pick up on figures, art and entertainment personalities, and even clothing styles.

  • Multidimensional analysis

    Analyzes videos from multiple dimensions, such as sound, actions, static images, and texts, to fully understand the video content so that you can get more thorough results.

Video Content Processing (VCP)

VCP performs intelligent video analysis and delivers video cover selection, video topics segmentation, video synopsis, and more capabilities. It can be used to extract attractive covers illustrating most of the video content, produce video abstract with video segments, and split news videos based on different topics.

  • Accurate topic segmentation

    Deep convolutional neural network (DCNN) is used to train and analyze massive volumes of video data for precise segmentation.

  • Key frame extraction

    Key frames are accurately extracted using the optical flow and combining time domain features based on video content understanding and structure analysis.

  • Stable and efficient

    Videos in various formats are supported, ensuring stable and efficient functions and effectively reducing costs.

Video Fingerprinting (VFP)

VFP adopts video fingerprinting technique to extract a video recording by its resultant "fingerprint". Characteristic components of the video are summarized with high stability to hinder video editing impacts, which enable that video to be uniquely identified.

VFP is of interest in many areas, including video similarity measurement, copyright protection, video moderation, advertisement monitoring, and digital rights management (DRM).

  • Precise identification

    Achieves 99% or higher accuracy in identifying similar videos regardless of the low quality or post-editing operations such as transcoding, frame rate, resolution, and ratio conversion, as well as adding text, logos and borders.

  • Video edit detection

    After comparing similar videos, VPF accurately detects the type of editing operations, such as mirroring a video, adding text, logos, and borders, and video clipping.

  • High robustness

    Provides database with hundreds of millions of fingerprinted video recordings, allowing millisecond-level search with 99.9% reliability.

  • Wide applications

    Offers all-inclusive technical solutions to satisfy diversified requirements in different industry fields, such as media, finance, education, and entertainment.

  • Video evidence in blocks

    Avoid video copyright infringement by tracing the source videos with the help of Blockchain Service (BCS) of HUAWEI CLOUD.

Video Content Moderation (VCM)

Video Content Moderation detects and filters out non-compliant content, such as pornographic elements, terrorism-related information, and violent content, greatly improving video review efficiency.

  • Accurate Review

    Uses Convolution Neural Network (CNN) algorithms and volumes of training samples to generate prediction models with high precision for real-time detection.

  • Easy to Use

    Provides RESTful APIs for easy use and quick integration into review systems.

  • Wide Application

    Detects inappropriate and terrorism-related content as well as advertisements to ensure video compliance.

  • Multidimentional Review

    Analyzes speech, actions, images, and text in videos to mitigate or eliminate non-compliance risks and improve the review efficiency.