In this paper we study ability to sparse the neural networks which are used in the task of person re-identification in multi-camera CCTV systems. Sparse neural network allows significant reducing of the computation complexity. It means increasing of processing speed for huge volume of data. The main idea of our research is decreasing of computation complexity with simultaneous preserving of neural network efficiency in the task of person re-identification. The paper consists of 4 parts and conclusion. The first part – introduction, describes sphere of person re-identification applying and the key problems in this field. The second part – related work, describes main state-of-the-art approaches related to person re-identification techniques. In the third part called proposed approach, we formulate technique of sparse neural network learning. The fourth part – experiment results, describes experiment conditions, constrains, training datasets, and results. In the conclusion we make our proposal on new technique usage.
S. Makov, A. Minaev, A. Nikitin, V. Voronin, E. Semenishchev, V. Marchuk, "Feature representation learning by sparse neural network for multi-camera person re-identification" in Proc. IS&T Int’l. Symp. on Electronic Imaging: Image Processing: Algorithms and Systems XV, 2017, pp 149 - 155, https://doi.org/10.2352/ISSN.2470-1173.2017.13.IPAS-211