Guardado en:
| Autores principales: | , |
|---|---|
| Formato: | Recurso digital |
| Lenguaje: | |
| Publicado: |
Zenodo
2026
|
| Acceso en línea: | https://doi.org/10.5281/zenodo.20210564 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866901950411309056 |
|---|---|
| author | Dr. Naveen N C Monisha H |
| author_facet | Dr. Naveen N C Monisha H |
| contents | <p><span class="fontstyle0">Abstract—</span><span class="fontstyle1">The situational awareness under<br>consideration requires the simultaneous situational<br>awareness in a number of threat dimensions, both the<br>behavioral conditions of individuals, and the physical<br>position of dangerous objects. I am going to present in<br>this paper one single unified and deep learning platform<br>named SurveilAI which can be used to perform real time<br>facial emotion recognition and weapon detection within<br>a single web-deployable surveillance platform. The two<br>perception tasks are reached by using single YOLOv8<br>models that are trained on domain-specific datasets. The<br>trained weapon detection module, based on YOLOv8s<br>and trained over 20 epochs on a Roboflow-curated<br>dataset of four weapon categories i.e., grenade, gun,<br>handgun and knife has a macro F1-score of 0.846 and an<br>overall mAP50 of 0.922. With a per-class recall value of<br>between 0.54 on contempt and 0.89 on sleepy, the<br>emotion recognition module, which is based on<br>YOLOv8n and trained on a nine-class dataset of facial<br>expressions that include anger, contempt, disgust, fear,<br>happiness, natural expression, sadness, sleepiness, and<br>surprise achieves an overall F1-score of 0.750 with a<br>maximally good confidence threshold of 0.371. Both<br>models have a web interface implemented based on Flask<br>with user authentication support, real-time image<br>inference and FLASH notification of entities detected<br>with bounding box overlays. Experimental testing<br>testifies that the YOLOv8 architecture will be highly<br>beneficial in terms of the simultaneous evaluation of<br>behavioral and physical threats during surveillance<br>systems, and the SurveilAI platform will offer a scalable<br>and practical base of intelligent security applications</span> </p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_20210564 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | SurveilAI: An Intelligent Surveillance Using Real-Time Facial Emotion Recognition and Weapon Detection Dr. Naveen N C Monisha H <p><span class="fontstyle0">Abstract—</span><span class="fontstyle1">The situational awareness under<br>consideration requires the simultaneous situational<br>awareness in a number of threat dimensions, both the<br>behavioral conditions of individuals, and the physical<br>position of dangerous objects. I am going to present in<br>this paper one single unified and deep learning platform<br>named SurveilAI which can be used to perform real time<br>facial emotion recognition and weapon detection within<br>a single web-deployable surveillance platform. The two<br>perception tasks are reached by using single YOLOv8<br>models that are trained on domain-specific datasets. The<br>trained weapon detection module, based on YOLOv8s<br>and trained over 20 epochs on a Roboflow-curated<br>dataset of four weapon categories i.e., grenade, gun,<br>handgun and knife has a macro F1-score of 0.846 and an<br>overall mAP50 of 0.922. With a per-class recall value of<br>between 0.54 on contempt and 0.89 on sleepy, the<br>emotion recognition module, which is based on<br>YOLOv8n and trained on a nine-class dataset of facial<br>expressions that include anger, contempt, disgust, fear,<br>happiness, natural expression, sadness, sleepiness, and<br>surprise achieves an overall F1-score of 0.750 with a<br>maximally good confidence threshold of 0.371. Both<br>models have a web interface implemented based on Flask<br>with user authentication support, real-time image<br>inference and FLASH notification of entities detected<br>with bounding box overlays. Experimental testing<br>testifies that the YOLOv8 architecture will be highly<br>beneficial in terms of the simultaneous evaluation of<br>behavioral and physical threats during surveillance<br>systems, and the SurveilAI platform will offer a scalable<br>and practical base of intelligent security applications</span> </p> |
| title | SurveilAI: An Intelligent Surveillance Using Real-Time Facial Emotion Recognition and Weapon Detection |
| url | https://doi.org/10.5281/zenodo.20210564 |