Real-Time Crowd Emotion Detector Using YOLO and FER

Facial expression analysis is an active area of computer vision, but analysing a crowded scene presents practical challenges: several faces may appear in one frame, faces may be partially occluded or small, and illumination and head pose can vary.This paper presents the design of a Real-Time Crowd Emotion Detector that combines YOLObased face detection with Facial Emotion Recognition (FER).The proposed pipeline accepts webcam or video input, extracts frames, detects multiple faces, crops the detected regions, predicts an emotion for each face, and aggregates predictions into a crowd-level emotion distribution and dominant category.The literature reviewed includes YOLObased facial expression detection, video-based YOLO and Vision Transformer classification, FER-YOLO-Mamba, EmoNeXt, and OpenFace 3.0.These works motivate efficient detection, expressive feature learning, and attention to practical robustness.The current project is at the system-design and implementation stage; therefore, this paper reports the proposed architecture and evaluation plan rather than measured accuracy or speed.Evaluation is planned using labelled images and videos, with emotion-classification metrics, face-detection measures where labels are available, processing frames per second (FPS), and qualitative analysis of difficult scenes.The system is intended as an expressionanalysis aid and does not claim to determine a person's internal emotional state with certainty.