Academic Journal
Human Detection in Surveillance Videos Based on Fine-Tuned MobileNetV2 for Effective Human Classification
| العنوان: | Human Detection in Surveillance Videos Based on Fine-Tuned MobileNetV2 for Effective Human Classification |
|---|---|
| المؤلفون: | Bouafia, Yassine, Guezouli, Larbi, Lakhlef, Hicham |
| المساهمون: | LaSTIC Laboratory, Université Mustapha Ben Boulaid de Batna 2, Heuristique et Diagnostic des Systèmes Complexes Compiègne (Heudiasyc), Université de Technologie de Compiègne (UTC)-Centre National de la Recherche Scientifique (CNRS) |
| المصدر: | ISSN: 2228-6179 ; Iranian Journal of Science and Technology, Transactions of Electrical Engineering ; https://hal.science/hal-04225289 ; Iranian Journal of Science and Technology, Transactions of Electrical Engineering, 2022, 46 (4), pp.971-988. ⟨10.1007/s40998-022-00512-6⟩. |
| بيانات النشر: | HAL CCSD |
| سنة النشر: | 2022 |
| المجموعة: | Université de Technologie de Compiègne: HAL |
| مصطلحات موضوعية: | Video surveillance, Human detection, Transfer learning, MobileNets, Convolution neural network, Human classification, [INFO]Computer Science [cs] |
| الوصف: | International audience ; With the high rate of accidents and crimes around the world, the importance of video surveillance is growing every day and intelligent surveillance systems are being developed to perform surveillance tasks automatically. Detecting human beings accurately in a visual surveillance system is crucial for diverse application areas. The first step in the detection process is to detect moving objects. Then, the moving object could be classified either in the human class or in the non-human class. Human classification is an important process to build effective surveillance system. In this article, an efficient human detection algorithm is proposed by processing the regions of interest (ROI) based on a foreground estimation. In our proposal, we used MobileNetV2 deep convolution neural network, designed to be used in embedded devices, with transfer learning approach to build fine-tuned model for an efficient classification of ROI into human or not human. We train the fine-tuned model on INRIA person dataset using three scenarios. The resulting models were extensively evaluated on INRIA test dataset benchmark and they achieved an F-Score value of 98.35%, 98.72%, and 98.90% which we consider very satisfactory performance. The best fine-tuned model was used for the classification stage which achieved an accuracy of 98.42%, recall of 99.47%, precision of 98.34% and F-Score of 98.90%. |
| نوع الوثيقة: | article in journal/newspaper |
| اللغة: | English |
| Relation: | hal-04225289; https://hal.science/hal-04225289 |
| DOI: | 10.1007/s40998-022-00512-6 |
| الاتاحة: | https://hal.science/hal-04225289 https://doi.org/10.1007/s40998-022-00512-6 |
| رقم الانضمام: | edsbas.9C8567CF |
| قاعدة البيانات: | BASE |
| DOI: | 10.1007/s40998-022-00512-6 |
|---|