Heterogeneous Video Ingestion to AI Models

The growing number of cameras in transport infrastructure and smart buildings requires solutions of the unified processing of heterogeneous video
inputs for Ai models.

This work analyzes streaming protocols suitable for real-time video processing and proposes a modular system for managing multi-camera streams, including the normalization of inputs into a unified format suitable for computer vision and artificial intelligence models. Part of the work
is the implementation of an experimental module for extracting frames with a timestamp, which is evaluated on a selected object detection model under conditions of concurrent processing of multiple streams.
‍