Abstract: Securing homes is a major challenge in a context where traditional key-, badge-, or code-based opening systems have intrinsic vulnerabilities: loss, copying, or unauthorized sharing. This work proposes the design of an intelligent residential access control system based on computer vision and facial recognition, intended for a family dwelling with three distinct access points: a back door, a veranda door, and a central door. The system is designed to recognize six pre-registered authorized individuals. The proposed approach relies on a hybrid architecture combining YOLO for fast face detection in images acquired by cameras installed at the three access points, a facial feature extraction module based on deep representations, and a decision mechanism incorporating a confidence threshold. After detection, facial regions are normalized and then submitted to a recognition module to identify the person. When the identity matches one of the six authorized individuals and the confidence level exceeds a predefined threshold, an opening recognized command is transmitted to the electromechanical mechanism of the concerned door. Otherwise, access is denied and a security event is recorded. The proposed evaluation relates to accuracy, precision, recall, F1-score, false positive rate, false rejection rate, and response time. The envisioned system thus constitutes a contactless, automated access control solution adaptable to multiple entry points of a single dwelling.

Keywords: artificial intelligence, facial recognition, YOLO, computer vision, access control, deep learning, residential security, liveness detection.

Title: Design of an intelligent residential access control system using YOLO-based facial recognition: application to a three-entry dwelling

Author: MBO NKETA OLGA BRAVO, OKITAKOY MANYA SYLVADOR, NEKA MUKENDI HENRY, MUKUNA WA MUKUNA FADER

International Journal of Novel Research in Computer Science and Software Engineering

ISSN 2394-7314

Vol. 13, Issue 3, September 2026 - December 2026

Page No: 26-35

Novelty Journals

Website: www.noveltyjournals.com

Published Date: 16-September-2026

Abstract: Urban security in Bandundu remains hampered by crime, theft, burglaries, and inadequate surveillance systems. This article uses a mixed-methods survey conducted with 103 residents and 20 institutional stakeholders to formalize the social acceptability of video surveillance. The theoretical model incorporates TAM, UTAUT, institutional trust, and privacy calculus. The results show a predominantly favorable acceptability, conditioned by feelings of insecurity, perceived usefulness, trust in authorities, and privacy guarantees. Age has no statistically significant relationship with acceptance of camera installation. χ^2=1,513The article proposes a formalization using logistic regression, path analysis p=0,982, ddl=7and structural equation modeling, followed by a BI architecture integrating data acquisition, storage, decision analysis, dashboards, and data security. Optimizing camera deployment is formulated as a maximum coverage problem under budgetary constraints. The main contribution lies in the articulation between social acceptability, statistical modeling, decision-making computing and digital governance in the Democratic Republic of Congo.

Keywords: video surveillance, social acceptability, TAM, UTAUT, logistic regression, structural equations, business intelligence, Bandundu, DRC.

Title: Social acceptability of urban video surveillance in Bandundu mathematical modeling, decision architecture and computer science challenges

Author: TABALA MBOMA Cypriennes

International Journal of Novel Research in Computer Science and Software Engineering

ISSN 2394-7314

Vol. 13, Issue 3, September 2026 - December 2026

Page No: 20-25

Novelty Journals

Website: www.noveltyjournals.com

Published Date: 16-September-2026

Abstract: Urban waste management in Kinshasa presents a major challenge, exacerbated by a sharp increase in rural exodus and growing waste production. This article proposes a deep learning method for the automatic detection and classification of waste into five categories: plastic, metal, glass, paper/cardboard, and organic. The methodology relies on the collection of 500 annotated real-world images and the generation of 2,500 synthetic images produced by WasteGAN, forming a dataset of 3,000 images distributed across 600 images per category. Two detection models, YOLOv8 and Faster R-CNN, were trained separately, and their results were then combined using the Weighted Boxes Fusion (WBF) method. This fusion method achieves the best performance, with an accuracy of 0.978 and a recall of 0.969, while YOLOv8 exhibits the best inference time (18 ms per image). The proposed approach provides a promising basis for an intelligent waste sorting system, aimed at improving urban sanitation in Kinshasa.

Keywords: urban waste, deep learning, WasteGAN , YOLOv8, Faster R-CNN, Weighted Boxes Fusion, object detection, classification, Kinshasa.

Title: Automatic detection and classification of urban waste in Kinshasa using deep learning: merging the YOLOv8 and Faster R-CNN models

Author: MBO NKETA OLGA BRAVO, NEKA MUKENDI HENRY, KUTUMBAKANA, MAVANGA SERGE, NZINGA EALE GUELORD, EKILA BOLA TRÉSOR

International Journal of Novel Research in Computer Science and Software Engineering

ISSN 2394-7314

Vol. 13, Issue 3, September 2026 - December 2026

Page No: 8-19

Novelty Journals

Website: www.noveltyjournals.com

Published Date: 16-September-2026

Abstract: The proliferation of dynamic networks, such as mobile ad hoc networks (MANETs) and vehicular networks (VANETs), poses major challenges to conventional routing protocols, which struggle to adapt to rapid topology changes and fluctuating quality of service. This paper presents an alternative based on multi-agent reinforcement learning (MARL). We propose MARL Routing, a fully distributed routing architecture where each node acts as an intelligent agent learning from local forwarding decisions via a cooperative Q- learning algorithm. The main contribution lies in the design of a composite reward function, a partial Q-table reset mechanism for increased responsiveness, and a sparse information-sharing protocol (Q gossip). Experimental validation, conducted under NS-3, compares MARL Routing to OLSR, AODV, and classical Q- routing. The results demonstrate significant superiority in highly mobile environments, with a delivery rate of 91% at 15 m/s compared to 60-70% for reference protocols, and a convergence time of less than 2 seconds after a sudden failure. Sensitivity analysis confirms the robustness of the chosen parameters.

Keywords: Adaptive routing, Dynamic networks, Multi-agent reinforcement learning, Distributed Q- learning, Mobility, Network protocols, MARL Routing.

Title: Adaptive Routing in Dynamic Networks by Multi-Agent Reinforcement Learning: Design, Modeling and Evaluation of MARL Routing

Author: MUKUNA WA MUKUNA FADER, EKILA BOLA TRÉSOR, MANZIA MANSANGA ELIANE

International Journal of Novel Research in Computer Science and Software Engineering

ISSN 2394-7314

Vol. 13, Issue 3, September 2026 - December 2026

Page No: 1-7

Novelty Journals

Website: www.noveltyjournals.com

Published Date: 16-September-2026