Sistem Deteksi Pelecehan Seksual Berbasis Deep learning dengan Mekanisme Early Warning, AutoRecording dan Pengiriman Bukti Terenkripsi Menggunakan AES-128 GCM

Arsyad, Arsyad Rizantha Maulana Salim (2026) Sistem Deteksi Pelecehan Seksual Berbasis Deep learning dengan Mekanisme Early Warning, AutoRecording dan Pengiriman Bukti Terenkripsi Menggunakan AES-128 GCM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027221049-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Pelecehan seksual merupakan masalah sosial serius yang sering tidak dilaporkan karena hambatan struktural dan kesulitan pembuktian hukum. Penelitian ini mengembangkan sistem deteksi otomatis berbasis Deep Learning pada kamera CCTV, menggunakan YOLO untuk deteksi subjek manusia serta CNN-LSTM dan DenseNet201 sebagai pembanding untuk klasifikasi aksi pelecehan fisik pada area vital tubuh berdasarkan fitur spasial dan temporal. Sistem mengintegrasikan mekanisme Early Warning melalui notifikasi Telegram, Auto-Recording yang menyimpan rekaman otomatis saat insiden terdeteksi, serta enkripsi ganda AES-128 GCM dan AES-256 GCM untuk menjaga kerahasiaan sekaligus integritas bukti digital. Sistem dibangun dengan TensorFlow/Keras dan antarmuka web berbasis Flask dan Socket.IO di atas arsitektur multiprocessing, menggunakan dataset 1.000 video pelecehan dan 1.000 video normal. Pada test set, CNN-LSTM mencapai akurasi 91,00% (AUC 0,9797) dan DenseNet201 90,00% (AUC 0,9734). Pada pengujian real-time menggunakan webcam dan CCTV, performa kedua model bersifat komplementer — CNN-LSTM konsisten unggul pada Precision sehingga menghasilkan lebih sedikit false alarm, sementara DenseNet201 unggul pada Recall, terutama pada kondisi kamera overhead. CNN-LSTM dipilih sebagai model utama karena Precision yang lebih tinggi, yang penting untuk validitas bukti hukum. Pengujian benchmark menunjukkan AES-128 GCM lebih cepat dengan throughput 857,65 MB/s, dengan integritas bukti dijamin melalui authentication tag 128-bit. Sistem ini diharapkan mampu mempercepat respons penanganan kejadian, melindungi korban, dan menyediakan alat bukti hukum yang sah untuk mendukung penegakan hukum.
=================================================================================================================================
Sexual harassment is a serious social issue that often goes unreported due to structural barriers and difficulties in legal proof. This research develops an automated Deep Learning-based detection system for CCTV cameras, using YOLO for human subject detection and CNN-LSTM and DenseNet201 as benchmarks for classifying physical harassment actions on vital body areas based on spatial and temporal features. The system integrates an Early Warning mechanism via Telegram notifications, an Auto-Recording feature that automatically saves recordings when an incident is detected, and dual AES-128 GCM and AES-256 GCM encryption to protect both the confidentiality and integrity of digital evidence. It was built with TensorFlow/Keras and a Flask and Socket.IO-based web interface on a multiprocessing architecture, using a dataset of 1,000 harassment videos and 1,000 normal videos. On the test set, CNN-LSTM achieved 91.00% accuracy (AUC 0.9797) and DenseNet201 90.00% (AUC 0.9734). In real-time testing using a webcam and CCTV, both models performed complementarily — CNN-LSTM consistently excelled in Precision, resulting in fewer false alarms, while DenseNet201 excelled in Recall, particularly under overhead camera conditions. CNN-LSTM was selected as the primary model due to its higher Precision, which is essential for the validity of legal evidence. Benchmark testing showed that AES-128 GCM is faster with a throughput of 857.65 MB/s, with evidence integrity guaranteed through a 128-bit authentication tag. This system is expected to accelerate incident response times, protect victims, and provide valid legal evidence to support law enforcement.

Item Type:	Thesis (Other)
Subjects:	T Technology > T Technology (General) > T57.62 Simulation
Divisions:	Faculty of Information and Communication Technology > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Arsyad Rizantha Maulana Salim
Date Deposited:	30 Jul 2026 03:02
Last Modified:	30 Jul 2026 03:02
URI:	http://repository.its.ac.id/id/eprint/139650

Actions (login required)

View Item