This article proposes a compact vectorized representation approach to realize reliable obstacle detection for autonomous ground vehicles. Bridging the gap between the advantage of occupancy grid map in general obstacle detection and its shortcomings in memory-consuming and computational cost, a novel obstacle representation by multiple convex polygons is first proposed. Specially, to overcome the computational challenge brought by many 3-D point cloud, based on ground plane extraction and nearest obstacle selection, 3-D obstacle points are transformed into 2-D obstacle points. In addition, the maintenance process of grid map with a fixed range is proposed to further improve efficiency. More importantly, the compact obstacle representation is realized via multiple convex polygons through double threshold-based boundary simplification and convex polygon segmentation. The compact vectorized representation is the main contribution, which achieves the goal of compactness and accuracy on the premise of ensuring effective and reliable obstacle detection. The proposed approach has been applied in a practical autonomous driving project because of its superior performance on general obstacles detection. In addition, the quantitative evaluation shows the superior performance on making use of fewer number of points (decreased by about 50%) to represent local static environments.