ABSTRACT
This chapter discusses unified vehicle detection, tracking, and trajectory prediction of objects around the ego vehicle and includes sections that explain individual tasks in detail. State-of-the-art methods for object detection based on deep learning can be broadly classified into two types: two-stage detectors and single-stage detectors. Although object detection has gained huge attention in autonomous driving applications, state of-the-art algorithms still have some limitations in realistic scenarios. Deep IDNet proposed a deformable part-based CNN for object detection by designing a deformation-pooling layer to learn the geometric deformations of all instances of a part. However, multiple-object tracking provides adequate support for autonomous driving applications. In the future, unified algorithms can be built using CNNs by sharing an encoder among all three tasks of autonomous driving. They demonstrated the importance of three crucial tasks—namely, object detection, tracking, and trajectory prediction.
