Lane Change Detection of Surrounding Vehicles in Dashcam Videos: A Synthesis of Methods and Challenges for Future Research

Abstract

Lane change events are a critical focus for road safety research. Detecting the lane-changing or cut-in behavior of surrounding vehicles using dashcam video has significant potential for supporting driver behavior monitoring and timely interventions during such events. However, as this field has not yet been systematically reviewed in a dedicated literature survey, researchers often face the challenge of manually filtering through many studies with overlapping keywords to identify relevant work. To address this gap, after investing significant efforts to seek the target studies of this specific domain, this paper presents a detailed review of each recent novel study from 2019. These existing approaches were also innovatively categorized into two main directions: direct model inference and logical inference based on lane marking. Each category is analyzed to highlight the shared characteristics and key differences, offering researchers a clearer understanding of the field's current landscape. Based on the analysis, several shared limitations specific to each direction were identified, and some open challenges that need to be solved by future research were proposed from both practical application and road safety perspectives. These include the heavy reliance on manually annotated data during preprocessing, the prevalent focus on evaluating algorithms only on event-specific video clips, and the lack of connection from detection methods to road safety research, among others. Addressing these issues is critical for advancing this field and strengthening its connection with real-world safety considerations