Any business dealing with large amounts of video data eventually comes to a point when human-based security monitoring and analysis is not enough. Especially when the number of video sources and stored video files grows exponentially. Eventually, this can become a tough challenge in terms of capabilities and resource usage. Having human operators means fluctuating levels of attention and inability to monitor and analyze multiple videos at a time, 24/7. However, we can easily solve this problem by applying highly precise, reliable and powerful AI solutions DeepX provides.
Smart security monitoring, archiving and interpretation system
Let’s break down the main functional parts of the system.
First of all, starting from the video sources, those are not limited to CCTV cameras. The system supports numerous stream formats – user webcams, rtsp, rtmp, WebRTC video streaming etc. The streams are ingested into AI processing module with the help of Gstreamer. Gstreamer pipelines are constructed depending on the incoming stream format.
Next, the ingested video stream goes to the AI image processing module containing CV/ML models. The models produce processed video and high level metadata on outputs. The video is archived by chunks of specified length, timestamped and uploaded to the media server, allowing for accessibility from the user interface. The produced metadata is written into a database (message cluster based on highly scalable Kafka). Video and metadata are synchronized by timestamps and the synchronization is frame-precise.
Finally, when the user wants to access the available videos and event highlights, such kinds of requests are handled by a corresponding module that communicates with database (message cluster), aggregates data and sends it to the user interface
From the user’s perspective, accessing an archived stream is simple. As soon as you enter the historical view section, you see a calendar containing highlights for the dates when the archiving was performed. Clicking the highlighted date gives the list of videos for each archived source sorted by timestamp. AI module has already processed them, and the database contains metadata produced by CV/ML models. In case the system has detected something during archiving a certain video chunk, it will highlight it in the list. This allows us to immediately access the videos by highlights where the AI system has detected something important.
In terms of CV/ML modules for AI processing, we can choose them when scheduling the stream source for archiving and processing. They can be tailored up to your business needs for specific purposes or chosen from the list of already available standard ones (like pose estimation, face or human detection etc.)
After we’ve chosen the processed video, we can view it in a player. The main feature of the player in historical view is the smart timeline. It contains the highlights of events the AI processing module has detected and allows us clicking and viewing them immediately, eliminating the need to view the whole video trying to spot something. The system has already done that job for us.
Hovering the mouse over the highlighted event also gives us the description, which CV/ML model has generated during stream processing.
Areas of application and benefits of AI-powered security monitoring for business
The solution for AI stream processing and viewing events in retrospective has numerous areas of application. Some of them are:
- CCTV security monitoring
- Patients monitoring in healthcare
- Logistics AI solutions
- Manufacturing (spotting defects)
- Any other sphere dealing with large amounts of video feed needing to be processed
Speaking about the benefits of AI for business, our solution includes the following:
- Fully automated security monitoring of multiple sources
- Custom AI processing with specifically tailored computer vision and machine learning models up to your needs
- High precision of CV/ML models outperforming human operators
- Stable operation (24/7 cctv monitoring)
- Video and metadata preserving, which allows for additional post-processing and analysis to be performed
- High scalability, allowing for processing of high and growing number of video sources
- Immediate access to the most important AI-spotted events