Diffused Color Exposure Balanced Live Stitching System

Abstract

The invention relates to real-time image processing, live video stitching, distributed edge computing, and professional media transmission architectures in multi-camera setups; specifically, it addresses the process of stitching high-resolution live video streams from multiple independent camera sources into 360-degree, Virtual Reality (VR), and spatial formats, where brightness, exposure, and color inconsistencies caused by cameras receiving light from different angles can be compensated at seamlines without causing live broadcast delay. This is achieved by integrating Image Signal Processors (ISPs), FPGAs, NPUs, or embedded microprocessors into camera heads or output ports to generate statistical color / brightness data (histograms) only for overlapping regions instead of the entire video frame.It is a distributed color / exposure compensation system used in low-latency broadcast workstations that enable the central server to perform only stitching and frequency-based blending (Multi-band Blending) by freeing it from the color optimization burden, and to correct the image hardware-wise during the transmission phase by extracting and calculating gain or 1D-LUT values ​​(luminance vectors) at the hardware level and feeding them back to the camera's own ISP or shader hardware, in order to avoid burdening the central processing unit with pixel analysis, and to perform distributed optimization by exchanging statistical packets over packet-switched lightweight protocols (P2P / UDP) between edge camera processors and to perform distributed optimization by freeing the central server from the color optimization burden in professional live broadcast, remote production (REMI) and stadium events conducted over terrestrial fiber or 5G&6G infrastructures, SMPTE ST 2110, SMPTE ST 2022-7, terrestrial fiber or 5G&6G infrastructures, and its feature is that the streams from the image sensors (3) synchronized with the timing layer (14),This involves converting the data packets (12) into only the overlapping areas via the cropping unit (5) and histogram engine (6); exchanging these lightweight packets between endpoints via telemetry ports (11) and calculating the local color coefficients by the controller (9); loading the coefficients to the hardware ISP (4) layer via the feedback interface (8) and hardware stabilizing the video at the output stage; and finally, the central server (15) producing the live VR output (20) with ultra-low latency by projection (17), seam tracking (18) and Laplacian blending (19) steps without carrying any pixel color optimization overhead.