Inter-Camera Person Tracking in Non-overlapping Networks
K. A. Shiva Kumar, K.R. Ramakrishnan, G. N. Rathna · 2017
Person tracking in non-overlapping camera networks is an important problem in many applications. Re-identification plays a significant role for successful tracking. Recently, much progress has been made in re-identification, but the traditional re-identification evaluation is not very suitable for tracking applications. It does not consider specific challenges of re-identification for person tracking such as open-set, dynamic and smaller gallery set and multiple cameras. We propose a new protocol for evaluating re-identification algorithms for person tracking considering issues as mentioned above. We also propose an on-line update scheme for improving the performance of re-identification algorithm. We learn a generic model of the persons off-line, then fine-tune the model to the specific person. We explain the generation of positive and negative samples. We have conducted experiments on six real-world datasets showing the effectiveness of proposed protocol and on-line update.