Deep Occlusion Reasoning for Multi-Camera Multi-Target Detection

People detection in single 2D images has improved greatly in recent years.However, comparatively little of this progress has percolated into multi-cameramulti-people tracking algorithms, whose performance still degrades severelywhen scenes become very crowded. In this work, we introduce a new architecturethat combines Convolutional Neural Nets and Conditional Random Fields toexplicitly model those ambiguities. One of its key ingredients are high-orderCRF terms that model potential occlusions and give our approach its robustnesseven when many people are present. Our model is trained end-to-end and we showthat it outperforms several state-of-art algorithms on challenging scenes.