CULS – INDOOR OCCUPANCY DETECTION DATASET

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F60460709%3A41110%2F23%3A96592" target="_blank" >RIV/60460709:41110/23:96592 - isvavai.cz</a>

Výsledek na webu

<a href="https://ap.pef.czu.cz/en/r-12193-conference-proceedings" target="_blank" >https://ap.pef.czu.cz/en/r-12193-conference-proceedings</a>

DOI - Digital Object Identifier

—

Jazyk výsledku

angličtina

Název v původním jazyce

CULS – INDOOR OCCUPANCY DETECTION DATASET

Popis výsledku v původním jazyce

A new dataset for occupancy detection in smart buildings such as universities is presented in this paper. The dataset can be used to train neural network models for this task (object recognition of person’s head). Detectable space in smart buildings is defined as corridors/common areas as well as, for example, classrooms and auditoriums. New dataset is specific and unique because it contains annotations of indoor occupants from three views: front, side and back. This is different from other datasets that normally focus on only one type of annotation. The dataset also considers the varied conditions that occur during detection – for example, the positioning of cameras in overhead, from the side, or other conditions, such as lighting. In the cooperation with Security Department of Czech University of Life Sciences Prague, the video recordings of five lecture rooms were obtained for a duration of ~372 hours, from which still images were created and all persons appearing there were manually annotated with bounding boxes. The number of these annotations amounts to 10 044 persons. Comparison was then made on this data with other publicly available datasets. Then, ResNet-50 model was trained using this dataset to determine if this dataset is applicable in machine learning. It was found that a similar dataset designed primarily to count people from different perspectives in auditoriums did not exist at the time of the research. Compared to other dataset, presented dataset is smaller in size, however by creating an experimental model based on ResNet50, it was found that in transfer learning, the model created is capable of inference and is therefore practically applicable. Hence, the dataset can be used in machine learning

Název v anglickém jazyce

CULS – INDOOR OCCUPANCY DETECTION DATASET

Popis výsledku anglicky

A new dataset for occupancy detection in smart buildings such as universities is presented in this paper. The dataset can be used to train neural network models for this task (object recognition of person’s head). Detectable space in smart buildings is defined as corridors/common areas as well as, for example, classrooms and auditoriums. New dataset is specific and unique because it contains annotations of indoor occupants from three views: front, side and back. This is different from other datasets that normally focus on only one type of annotation. The dataset also considers the varied conditions that occur during detection – for example, the positioning of cameras in overhead, from the side, or other conditions, such as lighting. In the cooperation with Security Department of Czech University of Life Sciences Prague, the video recordings of five lecture rooms were obtained for a duration of ~372 hours, from which still images were created and all persons appearing there were manually annotated with bounding boxes. The number of these annotations amounts to 10 044 persons. Comparison was then made on this data with other publicly available datasets. Then, ResNet-50 model was trained using this dataset to determine if this dataset is applicable in machine learning. It was found that a similar dataset designed primarily to count people from different perspectives in auditoriums did not exist at the time of the research. Compared to other dataset, presented dataset is smaller in size, however by creating an experimental model based on ResNet50, it was found that in transfer learning, the model created is capable of inference and is therefore practically applicable. Hence, the dataset can be used in machine learning

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

R - Projekt Ramcoveho programu EK

Rok uplatnění

2023

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Agrarian perspectives XXXII. Human Capital and Education in Agriculture

ISBN

978-80-213-3309-3

ISSN

2464-4781

e-ISSN

—

Počet stran výsledku

13

Strana od-do

143-155

Název nakladatele

PEF ČZU v Praze

Místo vydání

Praha

Místo konání akce

Praha

Datum konání akce

1. 1. 2023

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—