A Cascaded Ensemble of Sparse-and-Dense Dictionaries for Vehicle Detection

Vehicle detection as a special case of object has practical meaning but faces challenges, such the difficulty detecting vehicles various orientations, serious influence from occlusion, clutter background, etc. In addition, existing effective approaches, like deep-learning-based ones, demand large amount training time and data, which causes trouble for their application. this work, we propose dictionary-learning-based vehicle approach explicitly addresses these problems. Specifically, an ensemble sparse-and-dense dictionaries (ESDD) are learned through supervised low-rank decomposition; each pair (SDD) in is trained to represent either subcategory (corresponding certain orientation range or occlusion level) background cluster patterns) only gives good reconstructions samples corresponding subcategory, making ESDD capable classifying even though they exhibit appearances. We further organize into two-level cascade (CESDD) perform coarse-to-fine two-stage classification better performance computation reduction. The CESDD then coupled with downstream AdaBoost process generate robust classifications. proposed model used window classifier sliding-window scan over image pyramids produce multi-scale detections, adapted mean-shift-like non-maximum suppression adopted remove duplicate detections. Our evaluated on KITTI dataset compared other strong counterparts; experimental results effectiveness CESDD-based detection, demands small data.