3.0 0710  2026
Changed: Optimize Firesmoke detection with retraining the model 
Changed: Optimize uscapi and h5ssmain memory useage.
Changed: Optimize AI area rule and crossline engine.
Added: Support AI meta data storage on multi object storage part.
Added: Add person fall and fighting model and event.
Added: Add A8 AI accelerator on chip video decoding.
Added: Add AI event in web default view.
Added: Add fire lane blocking detection.

USC, based on a microservices architecture, can easily implement multi-node clusters. The following describes its usage:

By utilizing a multimodal large model, cross modal information retrieval applications between natural language and video images can be achieved. By inputting a sentence or a word, target images can be retrieved in seconds, making target retrieval wider, more accurate, faster, and easier. This allows security video backtracking to no longer be limited to traditional methods such as spatiotemporal and alarm retrieval, greatly improving the efficiency of target and event search, and making security business management more efficient and intelligent.

Recently, our project required integration with Avigilon ACC7. After some effort, we finally obtained Avigilon partners access. Avigilon Unity Video and ACC7 provide a RESTful API, but it seems somewhat limited, unable to retrieve low-latency video data. After some searching, we found Unity Video Media Gateway 8.6, but upon closer inspection, we discovered that this component is only used for transcoding and still cannot achieve low-latency video stream integration.