Journal influence
Higher Attestation Commission (VAK) - К1 quartile
Russian Science Citation Index (RSCI)
Bookmark
Next issue
№2
Publication date:
14 June 2026
Intelligent analysis video data to recognize car theft situations in a parking lot
Date of submission article: 23.07.2019
UDC: 004.93`1
The article was published in issue no. № 1, 2020 [ pp. 162-168 ]Abstract:Modern video surveillance systems are very popular to prevent and investigate various illegal inci-dents. The main disadvantage of traditional decisions in this area is that the operator directly assesses the current situation and makes a decision. With a large number of monitored cameras and due to the human factor, there is a high probability of increasing the time delay in determining the dangerous situ-ation. This leads to significant damage, especially in cases where it is necessary to respond promptly to an incident. Such cases include, for example, the situation of carjacking. These circumstances necessi-tate the introduction of systems for the intellectual analysis of video data. This paper submits a method for recognizing situations of car theft from parking based on stochastic grammars and deep neural networks. Recognition of an incident is at two main levels: at the lower lev-el, event recognition occurs, at the upper level, the situation as the most probable chain of events cor-responding to the grammar of car theft signatures. The analysis of the identified objects, their relative position, the dynamic characteristics of the trajectory of movement and the characteristics of the peo-ple’s postures, fills up the detection of possible hijacking events. The deep neural networks recognize the objects and people poses. These networks have a high degree of reliability. The article developed a simulation model of the system of recognition of the situations of theft of road transport, which based on the object detection module using a deep neural network. The history of certain objects in previous frames and, if necessary, data on the posture of a person improves the as-sessment of the reliability of event recognition. To describe the possible scenarios of car theft, the authors developed a stochastic grammar and cre-ated the test utility based on it. The test results of the developed method on the Mini-deone video da-taset showed its efficiency.
Аннотация:Для предотвращения и расследования различных инцидентов активно используются системы видеонаблюдения. Основным недостатком традиционных решений в этой области является то, что оценивает текущую ситуацию и принимает решение непосредственно оператор, поэтому при большом количестве контролируемых камер и в силу человеческого фактора существует высокая вероятность увеличения времени определения опасной ситуации. Это приводит к значительному ущербу, особенно в тех случаях, когда необходимо оперативно реагировать на инцидент, например, при угоне автомобильного транспорта. Данные обстоятельства обусловливают необходимость внедрения систем интеллектуального анализа видеоданных. В работе предложен метод распознавания ситуаций угона автомобильного транспорта с пар-ковки, основанный на стохастических грамматиках и глубоких нейронных сетях. Распознавание инцидента происходит на двух основных уровнях: на нижнем уровне распознаются события, а на верхнем – ситуации как наиболее вероятные цепочки событий, соответствующие грамматике сиг-натур угона автомобиля. Детектирование событий возможного угона выполняется на основе анализа выявленных объектов, их взаимного расположения, динамических характеристик траектории движения и особенностей поз людей. Распознавание объектов и поз людей осуществляется на основе глубоких нейронных сетей, характеризующихся на современном этапе развития высокой степенью достоверности. В статье описана разработанная имитационная модель системы распознавания ситуаций угона автомобильного транспорта, которая базируется на модуле распознавания объектов с помощью глубокой нейронной сети. Повышение оценки достоверности распознавания события осуществляется за счет учета истории определенных объектов на предыдущих кадрах и при необходимости данных о позе человека. Для описания возможных сценариев угона автомобильного транспорта разработана стохастическая грамматика, на основе которой создана тестовая утилита. Результаты тестирования разработанного метода на наборе данных Mini-deone video dataset показали ее работоспособность.
| Authors: A.Yu. Kruchinin (kruchinin-al@mail.ru) - Orenburg State University (Associate Professor), Orenburg, Russia, Ph.D, R.R. Galimov (rin-galimov@yandex.ru) - Orenburg State University (Associate Professor), Orenburg, Russia, Ph.D | |
| Keywords: hijacking recognition, deep neural network, stochastic grammars, event-situational approach, human posture recognition |
|
| Page views: 8561 |
PDF version article Full issue in PDF (4.91Mb) |
Интеллектуальный анализ видеоданных для распознавания ситуаций угона автомобиля на парковке
DOI: 10.15827/0236-235X.129.162-168
Date of submission article: 23.07.2019
UDC: 004.93`1
The article was published in issue no. № 1, 2020. [ pp. 162-168 ]
Modern video surveillance systems are very popular to prevent and investigate various illegal inci-dents. The main disadvantage of traditional decisions in this area is that the operator directly assesses the current situation and makes a decision. With a large number of monitored cameras and due to the human factor, there is a high probability of increasing the time delay in determining the dangerous situ-ation. This leads to significant damage, especially in cases where it is necessary to respond promptly to an incident. Such cases include, for example, the situation of carjacking. These circumstances necessi-tate the introduction of systems for the intellectual analysis of video data.
This paper submits a method for recognizing situations of car theft from parking based on stochastic grammars and deep neural networks. Recognition of an incident is at two main levels: at the lower lev-el, event recognition occurs, at the upper level, the situation as the most probable chain of events cor-responding to the grammar of car theft signatures. The analysis of the identified objects, their relative position, the dynamic characteristics of the trajectory of movement and the characteristics of the peo-ple’s postures, fills up the detection of possible hijacking events. The deep neural networks recognize the objects and people poses. These networks have a high degree of reliability.
The article developed a simulation model of the system of recognition of the situations of theft of road transport, which based on the object detection module using a deep neural network. The history of certain objects in previous frames and, if necessary, data on the posture of a person improves the as-sessment of the reliability of event recognition.
To describe the possible scenarios of car theft, the authors developed a stochastic grammar and cre-ated the test utility based on it. The test results of the developed method on the Mini-deone video da-taset showed its efficiency.
A.Yu. Kruchinin (kruchinin-al@mail.ru) - Orenburg State University (Associate Professor), Orenburg, Russia, Ph.D, R.R. Galimov (rin-galimov@yandex.ru) - Orenburg State University (Associate Professor), Orenburg, Russia, Ph.D
Ссылка скопирована!
| Permanent link: http://swsys.ru/index.php?page=article&id=4690&lang=en |
Print version Full issue in PDF (4.91Mb) |
| The article was published in issue no. № 1, 2020 [ pp. 162-168 ] |
The article was published in issue no. № 1, 2020. [ pp. 162-168 ]
Back to the list of articles
