Video surveillance plays an important role in Industrial Internet of Things (IIOT). The entire system is demanded to output useful surveillance information in a timely and accurate manner, thus having high requirements of enough computation and communication resources. In this work, we propose a mobile edge computing (MEC) based video surveillance network for face recognition applications, consisting of multiple camera sensors, relays, and the mobile edge server. Considering the limited computation and communication resources, we make joint decision for task offloading, wireless channel allocation, and image compression rate selection to obtain high average recognition accuracy and low average process delay. We adopt different image recognition algorithms for both camera sensors and the MEC server according to their own computation ability, and utilize the deep Q-network (DQN) algorithm as the decision-making module. Besides, we propose a two-layer hierarchical learning framework, i.e., DQN and layers based on back propagation neural network (DQN+NN) algorithm, to reduce the curse of dimensionality. Experimental results show that both the DQN and DQN+NN algorithms can efficiently handle multiple computation tasks with limited computation and communication resources in intelligence video surveillance scenarios, and the DQN+NN algorithm performs better in terms of the convergence and training efficiency.
All Science Journal Classification (ASJC) codes
- Computer Science (miscellaneous)
- Information Systems
- Human-Computer Interaction
- Computer Science Applications