UBOL: User-Behavior-aware One-Shot Learning for Safe Autonomous Driving

In autonomous and self-driving vehicles, visual perception of the driving environment plays a key role. To achieve this goal, the systems of the vehicle rely on deep neural networks which are extensively trained on manually annotated databases. This work proposes a semi-parametric approach and aims to bypass the manual annotation required for training vehicles perception systems used in autonomous and self-driving. We present a novel “User-Behavior aware One-Shot Learning (UBOL)” for efficient data generation, which takes user action and response into consideration in addition to the synthetic traffic data generated as Pareto optimal solutions from one-shot objects using a set of generalization functions. This adds more robustness and safety features to autonomous driving. The Pareto-optimization is applied to improve the performance of classification and ensure the distribution of target and generated data is identical. We evaluate the proposed framework on environment perception challenges encountered in autonomous vision.