High performance and fast object detection in road environments
Min-Sung Kang, Young-Chul Lim · 2017
In this paper, we present a high performance and fast object detection method based on a fully convolutional network (FCN) for advanced driver assistance systems (ADAS). Object detection methods based on deep learning have high performance but they require high computational complexity. Even if a method works on the high-performance graphics processing unit (GPU) hardware platform, it is hard to guarantee real-time processing. General object detectors based on deep learning try to localize too many classes of objects in various dynamic environments. The proposed detection method based on FCN improves detection performance and maintains real-time processing in road environments through various schemes related to the limitation of object class type, data augmentation, network architecture, and multi-ratio default boxes. Our experimental results show that the proposed method outperforms a previous method both in terms of performance and speed.generation AEB systems are not only able to detect vehicles ahead but also in more