This paper presents a comprehensive review of the application of federated learning (FL) in Intelligent Transportation Systems (ITS), focusing on three key scenarios: traffic flow prediction, traffic target recognition, and vehicular edge computing. ITS has become essential for managing urban traffic, improving transportation efficiency, and reducing environmental pollution. However, traditional centralized training approaches face challenges such as poor real-time performance, data silos, and privacy concerns. FL offers a promising solution by enabling decentralized model training while preserving data privacy. In traffic flow prediction, FL addresses the challenges of high privacy, rich data types, real-time requirements, and sensitive data structures. FL allows organizations to collaborate on model development without sharing raw data, reducing privacy risks. Recent studies have integrated FL with graph neural networks and blockchain to enhance security and efficiency. FL also supports personalized travel time estimation and crowd flow prediction, improving traffic management and route planning. In traffic target recognition, FL helps overcome data scarcity, high resource consumption, and computational load. FL enables distributed model training for road target recognition, including pothole detection and traffic sign recognition, while preserving data privacy. FL is also applied to in-vehicle driving detection, such as steering angle prediction and driver facial detection, improving safety and efficiency. In vehicular edge computing, FL supports decentralized model training and resource optimization. FL reduces communication overhead by compressing models and selecting optimal clients. FL also enhances resource efficiency by optimizing computation and storage. FL in vehicular edge computing is categorized into communication-efficient, resource-optimized, and security-enhanced approaches, each addressing specific challenges in FL. Overall, FL provides a secure and efficient solution for ITS applications, enabling collaborative learning while protecting user privacy. FL has shown promising results in various ITS scenarios, including traffic flow prediction, target recognition, and vehicular edge computing. Further research in FL is expected to contribute significantly to the development of ITS and the realization of safer, more efficient autonomous driving.This paper presents a comprehensive review of the application of federated learning (FL) in Intelligent Transportation Systems (ITS), focusing on three key scenarios: traffic flow prediction, traffic target recognition, and vehicular edge computing. ITS has become essential for managing urban traffic, improving transportation efficiency, and reducing environmental pollution. However, traditional centralized training approaches face challenges such as poor real-time performance, data silos, and privacy concerns. FL offers a promising solution by enabling decentralized model training while preserving data privacy. In traffic flow prediction, FL addresses the challenges of high privacy, rich data types, real-time requirements, and sensitive data structures. FL allows organizations to collaborate on model development without sharing raw data, reducing privacy risks. Recent studies have integrated FL with graph neural networks and blockchain to enhance security and efficiency. FL also supports personalized travel time estimation and crowd flow prediction, improving traffic management and route planning. In traffic target recognition, FL helps overcome data scarcity, high resource consumption, and computational load. FL enables distributed model training for road target recognition, including pothole detection and traffic sign recognition, while preserving data privacy. FL is also applied to in-vehicle driving detection, such as steering angle prediction and driver facial detection, improving safety and efficiency. In vehicular edge computing, FL supports decentralized model training and resource optimization. FL reduces communication overhead by compressing models and selecting optimal clients. FL also enhances resource efficiency by optimizing computation and storage. FL in vehicular edge computing is categorized into communication-efficient, resource-optimized, and security-enhanced approaches, each addressing specific challenges in FL. Overall, FL provides a secure and efficient solution for ITS applications, enabling collaborative learning while protecting user privacy. FL has shown promising results in various ITS scenarios, including traffic flow prediction, target recognition, and vehicular edge computing. Further research in FL is expected to contribute significantly to the development of ITS and the realization of safer, more efficient autonomous driving.