Recently, incremental learning for person re-identification receives increasing attention, which is considered a more practical setting in real-world applications. However, the existing works make the strong assumption that the cameras are fixed and the new-emerging data is class-disjoint from previous classes. In this paper, we focus on a new and more practical task, namely Camera Incremental person ReID (CIP-ReID). CIP-ReID requires ReID models to continuously learn informative representations without forgetting the previously learned ones only through the data from newly installed cameras. This is challenging as the new data only have local supervision in new cameras with no access to the old data due to privacy issues, and they may also contain persons seen by previous cameras. To address this problem, we propose a non-exemplar-based framework, named JPL-ReID. JPL-ReID first adopts a one-vs-all detector to discover persons who have been presented in previous cameras. To maintain learned representations, JPL-ReID utilizes a similarity distillation strategy with no previous training data available. Simultaneously, JPL-ReID is capable of learning new knowledge to improve the generalization ability using a Joint Plasticity Learning objective. The comprehensive experimental results on two datasets demonstrate that our proposed method significantly outperforms the comparative methods and can achieve state-of-the-art results with remarkable advantages.