This paper provides a comprehensive survey of advances in facial expression recognition (FER), covering methods, benchmarks, models, and datasets. It outlines the evolution of FER techniques from traditional methods to deep learning approaches, highlighting key challenges and solutions. The paper introduces a new taxonomy for FER, categorizing methods and datasets into controlled and uncontrolled environments, as well as images and sequences. It reviews existing deep learning models, including CNNs, RNNs, DBNs, and GANs, and discusses their applications in FER. The paper also presents a detailed analysis of popular FER datasets and suggests future research directions. It emphasizes the importance of robustness, adaptability, and handling uncertainty in real-world scenarios. The survey highlights the significance of FER in various fields, including computer vision, AI, and human-computer interaction, and discusses the growing market for emotion recognition. The paper also addresses challenges such as overfitting, limited training data, and expression-unrelated variations, and proposes solutions like data augmentation and preprocessing techniques. The survey concludes with recommendations for future research and development in FER.This paper provides a comprehensive survey of advances in facial expression recognition (FER), covering methods, benchmarks, models, and datasets. It outlines the evolution of FER techniques from traditional methods to deep learning approaches, highlighting key challenges and solutions. The paper introduces a new taxonomy for FER, categorizing methods and datasets into controlled and uncontrolled environments, as well as images and sequences. It reviews existing deep learning models, including CNNs, RNNs, DBNs, and GANs, and discusses their applications in FER. The paper also presents a detailed analysis of popular FER datasets and suggests future research directions. It emphasizes the importance of robustness, adaptability, and handling uncertainty in real-world scenarios. The survey highlights the significance of FER in various fields, including computer vision, AI, and human-computer interaction, and discusses the growing market for emotion recognition. The paper also addresses challenges such as overfitting, limited training data, and expression-unrelated variations, and proposes solutions like data augmentation and preprocessing techniques. The survey concludes with recommendations for future research and development in FER.