Maneuver Classification of Lane Change Based on Roadside Sensors Using Field Operational Test Data

Abstract

In this paper, a classification algorithm of lane change maneuver based on roadside sensors on highway is proposed. Data augmentation using field operational test data is also considered for scalability. The maneuver classification is composed of semantic maps and convolution neural network(CNN). The semantic map aims to represent a bird's eye view of both vehicle and road geometry, and the corresponding trajectory of the vehicle. The CNN is used to classify a lane change maneuver of multiple vehicles. While good performance of maneuver classification is shown with respect to a well-known dataset called highD, it is still necessary to consider scalability. Thus, the data augmentation is suggested to build a semantic map based on field operational test data. Despite different sensor characteristics of two datasets, it is demonstrated how the performance of CNN-based maneuver classification is improved in terms of scalability is demonstrated.