PIONEERING THE USE OF AI IN VIDEO COMMUNICATIONS: REAL-TIME INAPPROPRIATE CONTENT DETECTION

In the era of widespread video streaming platforms, ensuring safe and appropriate content in real-time is paramount, especially for audiences including adolescents.This paper presents a novel AI-integrated system for real-time detection of inappropriate content-such as nudity, obscene behavior, and profanity-in live video communications.Built upon the Pyjam streaming platform, our implementation leverages optimized WebRTC for low-latency transmission, lightweight AI models (e.g., MobileNet via ONNX for visual detection and Vosk for speech-to-text profanity filtering), and hardware acceleration techniques to achieve end-to-end latencies as low as 20 ms.We detail the system architecture, code improvements for latency reduction, and evaluation results, demonstrating superior performance in content moderation while maintaining user privacy and scalability.Our approach pioneers the fusion of edge computing and distributed AI, offering a blueprint for future video platforms.The system not only interrupts broadcasts upon detection but also applies dynamic actions like blurring or blackouts, ensuring compliance with child safety standards.