A reconfigurable smart camera implementation for jet flames characterization based on an optimized segmentation model

Gerardo Valente Vazquez-Garcia, Carmina Perez Guerrero, Eduardo Garduño, Miguel Gonzalez-Mendoza, Adriana Palacios, Gerardo Rodriguez-Hernandez, Vahid Foroughi, Alba Àgueda, Elsa Pastor, Gilberto Ochoa-Ruiz · 2026

Read A reconfigurable smart camera implementation for jet flames characterization based on an optimized segmentation model by Gerardo Valente Vazquez-Garcia, Carmina Perez Guerrero, Eduardo Garduño, Miguel Gonzalez-Mendoza, Adriana Palacios, Gerardo Rodriguez-Hernandez, Vahid Foroughi, Alba Àgueda, Elsa Pastor, Gilberto Ochoa-Ruiz (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-03267.txt.

Introduction

Vazquez-Garcia and colleagues build a smart camera platform that segments and characterizes jet flames on the device for industrial fire safety. Running an optimized UNet model on a system-on-chip FPGA, the platform processes video at the edge with low latency instead of shipping it elsewhere. Shrinking the model from 7.5 million to 59,095 parameters, the system reaches 30 frames per second without losing accuracy.

Catalog record

CREATORGerardo Valente Vazquez-Garcia, Carmina Perez Guerrero, Eduardo Garduño, Miguel Gonzalez-Mendoza, Adriana Palacios, Gerardo Rodriguez-Hernandez, Vahid Foroughi, Alba Àgueda, Elsa Pastor, Gilberto Ochoa-Ruiz
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE67827
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.03267

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A reconfigurable smart camera implementation for jet flames characterization based on an optimized segmentation model Gerardo Valente Vazquez-Garcia, Carmina Perez Guerrero, Eduardo Garduño, Miguel Gonzalez-Mendoza, Adriana Palacios, Gerardo Rodriguez-Hernandez, Vahid Foroughi, Alba Àgueda, Elsa Pastor, Gilberto Ochoa-Ruiz Abstract In this work we present a novel framework for fire safety management in industrial settings through the implementation of a smart camera platform for jet flames characterization. The proposed approach seeks to alleviate the lack of real-time solutions for industrial early fire segmentation and characterization. As a case study, we demonstrate how a System-on-Chip (SoC) Field Programmable Gate Arrays (FPGAs) -based platform, running optimized Artificial Intelligence (AI) models can be leveraged to implement a full edge processing pipeline for jet flames analysis. Jet flames are one of the less studied sources of major accidents. In this paper we extend previous work on computer-vision jet fire segmentation by creating a novel experimental set-up and system implementation for addressing this issue, which can be replicated to other fire safety applications. The proposed platform is designed to carry out image processing tasks in real-time and on device, reducing video processing overheads, and thus the overall latency. This is achieved by optimizing a UNet segmentation model to make it amenable for an SoC FPGAs implementation; the optimized model can then be efficiently mapped onto the SoC reconfigurable logic for massively parallel execution. For our experiments, we have chosen the Ultra96 platform, as it also provides the means for implementing full-fledged intelligent systems using the SoC peripherals, as well as other Operating System (OS) capabilities (i.e., multi-threading) for systems management. For optimizing the model we made use of the Vitis (Xilinx) framework, which enabled us to optimize the full precision model from 7.5 million parameters to 59,095 parameters (125x less), which translated into a reduction of the processing latency of 2.9x. Further optimization (multi-threading and batch normalization) led to an improvement of 7.5x in terms of latency, yielding a performance of 30 Frames Per Second (FPS) without sacrificing accuracy in terms of the evaluated metrics (Dice Score). We also show that the proposed architecture and model can effectively produce the desired characterization of horizontal jet flames, attaining performance similar to that of the full precision model. 1 Introduction Figure 1 : Different types of flames patterns in industrial settings: pool, ball and jet fires. Modified from Palacios (2011). Figure 2 : Visual example of jet fire characterization based on segmentation of three temperature zones. The production, handling and transport of hazardous materials in the chemical industry have increased during the last decades leading to a higher frequency of major accidents. The consequences that these have had on people and the environment have highlighted the need to improve industrial safety measures. This has been reflected in the implementation of legislation such as the Seveso Guideline on the prevention of major accidents in certain industrial activities 32 . Thus, the development of research studies on the main characteristics, mathematical modeling, and estimation of consequences of these major accidents occurring in the chemical industry around the world is of paramount importance. Fires are one of the most common and serious threats to the safety of both equipment and people operating in process units. Among the various types of major fires that can occur (see Fig. 1 ), jet fires (Fig. 2 (a) and (b)), are one of the lesser-known types, something that has changed recently, with a surging interest in the research community due to safety concerns for high-pressure storage and transport of fuels like hydrogen. Jet fires are a particular type of serious major fire accident that can occur both in fixed installations (process plants, storage, or pipes) and, less frequently, during transport (trucks or tank cars). Although jet fires occur with some frequency, many of them are not reported, since they are often accompanied by other events of greater magnitude, such as explosions. Jet fires can be evaluated based on specific geometrical features (Fig. 2 (c) and (d)). This is relevant from a risk management perspective because the probability of domino effect increases due to flame impingement, for example. Also, being able to predict the shape and proportions of a jet fire is relevant since it is possible then to identify the optimal space for the location of equipment and structures. A great majority of consequence analyses regarding jet fires, that evaluate both geometric features and radiation loads, employ physics-based models, such as Computational Fluid Dynamics (CFD) 10 , or empirical models, which are based on correlations developed from a set of experiments. However, few works have tackled this problem from a data-centric perspective. Furthermore, most empirical models use a simplified flame geometry and a fixed emissive power to determine the thermal radiation load on a target. Research done so far to characterize the main geometrical features of jet flames include the work from 11 , which presents a systematic study of the factors that affect the lift-off distance, which is the length between the burner exit and the base of the lifted flame. It was found that the lift-off distance varies linearly according to the exit velocity, and it is independent of the burner diameter. 35 also analyzed the shape, length, and width of relatively large vertical jet fires at sonic and subsonic exit velocities. The flame boundary was defined at a temperature of 800 K, and the results indicated that a cylindrical shape could accurately describe the shape, length, and diameter of the flame. Additionally, 7 introduced an extensive review and re-thinking of jet flame heights and structure, including a proposal of dimensionless correlations for the atmospheric jet flames’ heights and lift-off distances. It was found that the same flow rate parameter could be used to correlate both heights and flame lift-off distances, and based on that, an equation was proposed to predict jet flame height including several fuels and orientations. More recently, there has been a surging interest in using Machine Learning (ML) techniques to model jet fires. For instance, 25 presented an overview of ML techniques for low-cost and high-fidelity fire predictions where it was found that ML has the capacity to provide full-field predictions faster than CFD simulations. In this sense, a recent research work ( 39 ) has explored the use of Deep Learning (DL) -based computer vision methods for jet fires segmentation and characterization. Those methods were conceived for offline applications and made use of models deployed in cloud applications (i.e., they were not optimized for edge applications such as smart sensors). However, since monitoring jet fires is a critical task to provide a timely and adequate response, ML models must be capable of detecting and characterizing such situations preferably as smart sensors providing real-time alerts. This entails that these devices should be in close proximity to the areas where jet fires can potentially occur. As a solution for video processing in industrial Internet of Things (IoT) applications, several researchers and practitioners have proposed the use of smart cameras ( 33 ) , which main purpose is to implement all the functionalities of a conventional camera system (i.e., image acquisition, storage and video outputting) with an important difference: such smart devices are capable of performing image processing at the edge. The

Read the complete text