Flexible solution for modeling and tracking generic dynamic 3D environments

The traffic environment is a dynamic and complex 3D scene, which needs accurate models to represent it and reliable algorithms to perceive it. This paper presents a new model for representing the dynamic 3D environment of the traffic scene, the particle based dynamic elevation map enhanced with gray level information. A new tracking algorithm, based on the measurement cues extracted from dense stereovision, is employed for estimating the state of the proposed model. The multimodal probability density of a map cell´s state concerning the cell´s speed, height and gray level, is approximated by a population of particles, which can migrate from one cell to another using their speeds. The particle population is updated using a weighting-resampling mechanism based on the particle´s fitness with the measurement data, which consists of the raw heights obtained from stereovision and the pixels of the grayscale left image. The measurement model used for particle weighting is designed by taking into account the specifics of the stereovision sensor. The dynamic elevation map tracking system is able to provide a dense and accurate representation of the observed scene, to improve the density and accuracy of the stereo-based environment perception, and to enhance it with dynamic information. The final result is an accurate virtual representation of the perceived 3D scene.