Segmentation of Driving Areas for Autonomous Vehicle based on Deep Learning Method Using Automotive Camera Senosor
Si-Hyeon Lee, Yeonsik Kang · Journal of Institute of Control Robotics and Systems · 2020
This paper employs a deep learning method for segmenting drivable and non-drivable areas in an urban environment using the BDD (Berkeley Deep Drive) 100K image database. In particular, we propose a semantic segmentation network using an encoder module based on depth-wise separable, non-bottleneck-1D, and pyramid pooling. The BDD 100K dataset provides reference classification results of directly drivable area, alternative drivable area, and none, under different weather, time, and location conditions. The performance of the developed network is verified using a computation environment equipped with a GPU (Graphics Processing Unit) for evaluating its feasibility of real-time implementation.