Environment perception using dynamic polylines and particle based occupancy grids

Modeling and tracking dynamic entities in the driving environment is a complex task, as one has to accommodate multiple types of scenarios. Extraction of dynamic properties of obstacles becomes difficult when the measurement sensors do not provide speed directly. The dynamic polyline representation of obstacles is a compromise between the rigid model-based cuboid representation and the model-free representation of occupancy grids. This paper presents a polyline-based obstacle extraction system, based on a particle-based occupancy grid generated by processing of stereovision data. We have developed a real-time flexible method for occupancy grid modeling and representation using particles that move from cell to cell and are created and destroyed based on measurements derived from stereovision. The results of the occupancy grid processing are subjected to the Border Scanner algorithm, which extracts polylines from occupied cells, by taking into consideration only the most relevant information of the grid. The resulting system is able to extract individual objects from the occupancy grid, model them as polylines, and estimate their speed without making assumptions about their shape or size.