Driver assistant systems enhance traffic safety and\r\nefficiency. The accurate 3D pose of a front vehicle can help\r\na driver to make the right decision on the road. We\r\npropose a novel real-time system to estimate the 3D pose of\r\nthe front vehicle. This system consists of two parallel\r\nthreads: vehicle rear tracking and mapping. The vehicle\r\nrear is first identified in the video captured by an onboard\r\ncamera, after license plate localization and foreground\r\nextraction. The 3D pose estimation technique is then\r\nemployed with respect to the extracted vehicle rear. Most\r\ncurrent 3D pose estimation techniques need prior models\r\nor a stereo initialization with user cooperation. It is\r\nextremely difficult to obtain prior models due to the\r\nvarying appearance of vehicles� rears. Moreover, it is\r\nunsafe to ask for drivers� cooperation when a vehicle is\r\nrunning. In our system, two initial keyframes for stereo\r\nalgorithms are automatically extracted by vehicle rear\r\ndetection and tracking. Map points are defined as a\r\ncollection of point features extracted from the vehicle�s rear\r\nwith their 3D information. These map points are inferences\r\nthat relate the 2D features detected in following vehicles�\r\nrears with the 3D world. The relative 3D pose of the\r\nonboard camera to the front vehicle rear is then estimated\r\nthrough matching the map points with point features\r\ndetected on the front vehicle rear. We demonstrate the\r\ncapabilities of our system by testing on real-time and\r\nsynthesized videos. In order to make the experimental\r\nanalysis visible, we demonstrated an estimated 3D pose\r\nthrough augmented reality, which needs accurate and\r\nreal-time 3D pose estimation.
Loading....