A 2.5D grids based moving detection methodology for autonomous vehicle

摘要

A reliable perception of the real world is the key component in driver assistance systems. In this paper, a novel 2.5D grids based moving objects detection method is proposed. First, we construct a predicted local grid stack to store the historical filtered observations and the current observation. By merging the observations, a background model can be obtained. Then we detect and track moving objects through comparing the background model with the current local grid. Finally, we remove the moving cells in the current local grid to obtain the filtered observation and update the stack by replacing the current observation with the filtered observation. The experimental results based on the KITTI public dataset show the high detection accuracy and robustness of the proposed method.