① Video search using natural languages

When searching for a specific scene from surveillance camera footage, it was traditionally necessary to check the footage for each camera and timeframe to find the relevant scene. As the recording period and the number of cameras increased, it took more time to find the necessary information. VSS has a function where an AI agent analyzes content entered in natural language and performs searches based on the characteristics of events, people, and objects in the video. Users can perform searches using language similar to everyday conversation without having to set detailed search conditions. Furthermore, in addition to simple keyword searches, the AI analyzes the entered content and automatically selects the optimal search methodaccording to the search content, such as searches targeting actions and events (EmbedSearch)​,searchestargeting the appearance and attributes of people and objects (AttributeSearch), or Fusion Search which combines these. Therefore, users can perform video searches tailored to their purpose without having to be aware of differences in search methods. In addition to the relevant video clips, the search results also display the reasoning behind the answer, such as the detection time and similarity. It also supports refined searches by specifying a particular period or video source, allowing users to efficiently extract the desired scene from a large amount of recorded video.

As shown in the diagram above, video searches using natural languages can be performed following the steps below.

1. Entering search criteria: Enter the criteria for the target or event you want to search for using natural language.
2. AI-generated answer (summary): The AI analyzes the search results and displays a summary of the answer to the question in natural language.
3. Check the basis for the answer: You can check the start and end times of each scene, as well as the basis for the AI's judgment (clothing, actions, etc.).
4. List of relevant scenes: Scenes related to the question are extracted and displayed as thumbnails. Clicking the play button will play the video of the relevant section.

② Summarizing long-duration videos and generating reports

Because surveillance cameras continuously record video 24 hours a day, 365 days a year, it takes a considerable amount of time to understand the entire content of the footage. Even just checking for anomalies requires reviewing long hours of recorded video, which places a significant burden on the monitoring staff. VSS utilizes generated AI to analyze the entire video and has the function of summarizing the events and important occurrences that occurred in natural language. Users can grasp the content of the entire video in a short time without having to review the entire long hours of footage. Furthermore, the summary results are organized not only into text but also into a report that includes the time of occurrence and representative images for each event. This allows users to efficiently access the necessary information while visually confirming "when, where, and what happened."

As shown in the diagram above, summarizing long-duration videos and generating reports can be done in the following way.

1. Adding video: Upload the video you want to summarize, or add an RTSP stream.
2. Video Selection: Review the content from the thumbnails and select the video you want to summarize.
3. Summary Instructions&Condition Input: Enter summary instructions and conditions in natural language.
4. Summary of Summarization Results: This section displays the objects and prompts used during summarization, as well as the summary generated by the AI.
5. Report Output: You can download a summary report in PDF or Markdown format.
6. Video Playback: You can play the video that served as the basis for the summary to check the details.

Furthermore, the generated summary results can be output as reports inPDForMarkdown format. The reports include not only the summary content but also chronological event information and snapshots, making them directly usable as monitoring results reports or incident records. This eliminates the need for personnel reviewing the video to manually organize the content, leading to more efficient information sharing and reporting.

③ Questions and answers regarding the video

When reviewing surveillance footage to find necessary information, it may be necessary to rewatch the footage and verify its contents even after searching for the target scene. This is especially true for long videos containing multiple events, where simply organizing the necessary information can be time-consuming.

VSS has the ability to answer questions about video in natural language through dialogue with an AI agent. Users simply enter their questions in a chat format, and the AI generates answers based on video analysis results and event information. There is no need to learn specialized search criteria or operations, and users can ask exactly what they want to know, making video review a more intuitive process.

Furthermore, a key feature of this function is that it doesn't simply display search results, but generates answers based on the content of the video. By utilizing the information obtained through searches and summaries, users can engage in a dialogue-based discussion about events and objects that occurred in the video, allowing for efficient access to necessary information.

Furthermore, since questions and answers are stored chronologically in a chat format, it's possible to ask additional questions based on previous answers, rather than ending with just one question. A major feature of this function is that you can gradually understand the situation through dialogue without having to rewatch the video multiple times.

As shown in the diagram above, the question-and-answer session regarding the video can be carried out in the following manner.

1. Adding a video: Upload the video you want to ask about.
2. Video Selection: Check the content from the thumbnail and select the video you want to ask about.
3. Enter your question: Enter your question about the video in natural language. You can also ask follow-up questions to the AI's answer.
4. Question display: Questions entered by the user will be displayed in a chat format.
5. Display of Answers: The AI's answers to user questions are generated and displayed based on video analysis results and event information.

④ Alert notifications based on conditions

It's not practical for people to constantly monitor surveillance camera footage, and there's a risk of missing anomalies or important events. Therefore, a system is needed to notify users at the appropriate time, enabling a swift response.
VSS analyzes video in real time and has a function to notify the user of an alert when it detects an event that matches pre-set conditions. By automatically monitoring the occurrence of events, users can understand the situation when an important event occurs without having to constantly check the video. On the execution screen, the time of occurrence, the video in question, and related snapshots are organized and displayed for each detected event. Since the relevant video can be accessed directly from the notified alert, it is possible to smoothly carry out everything from checking the situation to investigating the cause.

Furthermore, since the generated alerts are managed as a history, they can be used not only for real-time notifications but also for reviewing past events and checking the status of responses. A key feature of this function is that it supports continuous event management, including not only monitoring but also equipment management and operational records.

As shown in the diagram above, condition-based alert notifications can be executed in the following manner.

1. Creating an alert: Create an alert by entering the content of the alert you want to be notified about in natural language.
2. Alert Summary: An overview of the alert created by the AI (target stream, detection details, status, etc.) is displayed.
3. Real-time notifications: When an event matching the alert criteria is detected, an alert is generated immediately.
4. History List: You can view a list of alerts that have been generated in the past. You can also view the time of generation, the type of alert, and even the video footage from the time of generation via the thumbnail.

As described above, VSS 3.2.1 offers four agent workflows: searching for video, understanding the entire video, interacting with video, and detecting events in real time. These functions can be used independently, but by combining them, the entire process from video search to situation assessment, reporting, and monitoring can be streamlined. Users can select the optimal workflow according to their purpose, enabling them to introduce AI-powered video analysis into various tasks such as monitoring, facility management, and incident response.