3D Object Detection and Tracking Methods using Deep Learning for Computer Vision Applications

Abstract

3D multi-object detection and tracking is an essential constituent for many applications in today's world. Object detection is a technology related to computer vision and image processing that allows us to detect instances of certain classes. There are numerous applications like robotics, autonomous driving and augmented reality. A bounding box often defines the region of interest and then is used to classify into respective categories. Due to identical appearance and shape of various objects and the interference of lighting and shielding, object detection has always been a challenging problem in computer vision. Conventional 2D object detection yields four degrees of freedom axis-aligned bounding boxes with centre (x, y) and 2D size (w, h), the 3D bounding boxes generally have 6 Degrees of freedom: 3D physical size (w, h, l), 3D centre location (x, y, z). 2D object detection and tracking methods do not provide depth information to perform essential tasks in various computer vision applications. One among them is the Autonomous driving. 3D object detection includes depth information that provides more information on the structure of detected object. More information is required to make decisions accurately in different fields where 3D object detection and tracking can be applied. In this paper various 3D object detection and tracking methods are elaborated for various computer vision applications, this includes various fields such as robotics, driving, space field and also in the military.