Person Re-Identification by Contour Sketch Under Moderate Clothing Change

Person re-identification (re-id), the process of matching pedestrian images across different camera views, is an important task in visual surveillance. Substantial development re-id has recently been observed, and majority existing models are largely dependent on color appearance assume that pedestrians do not change their clothes views. This limitation, however, can be issue for when tracking a person at places time if (e.g., criminal suspect) changes his/her clothes, causing most methods to fail, since they heavily relying thus inclined match another wearing similar clothes. In this work, we call under clothing "cross-clothes re-id". particular, consider case only his moderately as first attempt solving problem based visible light images; wears thickness, shape would significantly weather does substantially within short period time. We perform cross-clothes contour sketch image take advantage human body instead information extracting features robust moderate change. Due lack large-scale dataset re-id, contribute new consists 33698 from 221 identities. Our experiments illustrate challenges demonstrate effectiveness our proposed method.