Video surveillance using deep transfer learning and deep domain adaptation: Towards better generalization