CCTV Cameras have been augmented with Artificial Intelligence (AI) capabilities like object detection, and scene classification and alerting for quite some time now. The CCTV or surveillance security market is growing at a cumulative annual growth rate (CAGR) of 16.8% this decade alone, according to Fortune Business Insights.
As the market expands and adoption grows without significant improvement in technology (since the early integrations with computer vision) or better user experience, problems like technical usage barriers (personal installation and maintenance issues), and the time-consuming nature and difficulty of security reviews and search for a particular event or incident begin to emerge.
Surveillance.chat has combined some of the advanced foundational AI models, fine-tuned for surveillance to give everyday users the ability to query their surveillance cameras in a chat format, reducing the technicalities of operating surveillance systems while also adding convenience to security reviews, making things as simple as chatting with a friend and asking them questions.
Multi-modal Large Language Models that started gaining popularity and practical usage towards the end of last year are now advancing at a very fast rate. The top AI companies are now developing frontier models that include visual and audio, alongside text embeddings from the get-go. Meta AI recently released V-JEPA 2, a joint-embedding architecture for visual learning inorder to develop a world model, which will come in handy for motion and incident forecasting in surveillance.
“As much as the solution we are building is an AI solution, the major problem we are tackling is an engineering problem,” stated Surveillance.chat’s founder Aliyu Daku. “We are finding ways to efficiently implement video search and video indexing across 24 hours a day, 7 days a week continuous stream from multiple CCTV cameras. Our system has to optimally feed the models visual data in a way that they align multiple timestamps and not miss anything, while only analysing the absolutely necessary amount of footage for efficiency” he concluded.
For the team to be able to face and tackle these challenges, their collective past experiences have been of help. In the early 2010s, most of the team was responsible for one of Nigeria’s early indigenously built video streaming platforms, AfricaMars, later renamed 9flix TV. Building which, the team handled problems of streaming efficiency, resumable video uploading and optimization, and search and recommendation systems even in those early days of machine learning.
About Surveillance.Chat
Surveillance.chat is an AI solutions provider working to provide accurate visual analysis for security, business and everyday tasks. The company has sprouted out of AI development efforts behind several other projects. The Surveillance.chat team has also been working on Object-Centric Hierarchical Embedding with Multimodal Projections (OHEMP), which is a novel approach to token embedding for grounded representation learning.
