The Research On Dynamic Vision Sensor Networks

The nodes sensing model in VSN is called directional sensing model, and the node view is limited. So the deployed nodes should not be static. Nodes are generally divided into the following three types: removable, rotating, and both of the two. For this network, we call it Dynamic Vision Sensor Networks (DVSNs). It combines the vision sensor network and mobile sensor network, with greater research and application value.Study of DVSNs is generally focused on network architecture, vision sensor node, network coverage optimization, visual information processing and quality of service. The paper mainly researches localization in DVSNs, network coverage optimization and other related issues.Network localization is not only the base of sensor network, but also the premise of covering algorithm implementation. This paper describes the current situation of localization algorithm, describes in detail some typical algorithm and studies APTI algorithm based on sub-step refinement. Based on the thought of sub-step refinement, this paper introduced Aitken iteration into localization algorithm. The algorithm transferred the problem into the problem of seeking minimum error, then the optimal solution get from iterative algorithm is the final optimal positioning solution.The coverage problem of DVSNs can be divided into three aspects: area coverage, point target coverage and large object coverage. In this paper, the three coverage issues are discussed.For area coverage problem, we studied self-organization control algorithm of two-step virtual potential field force. The algorithm takes advantage of the interaction between vision sensor nodes'virtual attraction and repulsion makes the node move in the regions along the moving force until the force get balance. Meanwhile the vision sensor nodes randomly arranged in the original area will be uniformly distributed in the monitoring area. Then by using virtual potential field force algorithm again for the sensor node's"centroid"and adjust the direction of node, the coverage redundancy can be eliminated and enhance the effect of network coverage.For point target coverage, the paper proposes two coverage optimization algorithms. The first method is aimed at direction adjustable sensor model, to adjust visual sensor nodes'position and direction by using genetic algorithm's optimization advantages. Since optimization is for multi-objective optimization, the genetic algorithm based on weighted is applied. The simulation algorithm shows with the increase of genetic algebra, network target coverage rate will be greatly increased. Another method brings in greedy algorithm's ideas, establishing coverage relationship table, the node with widest coverage gets priority. Then update the coverage relationship, remove the working sensor and covered targets from the table. Repeat greedy algorithm until all the nodes is working every target is monitored. But when the widest covered node is not unique, if choose node randomly, coverage effect will be influenced. So"contribution rate"model is brought in, it makes the node with greatest contribution win the priority to work. Simulation results show that this algorithm can optimize network coverage rate as well as solve the problem of nodes'coverage conflict.Large object coverage is specific to DVSNs. When using vision node, we usually want to get object color, size and other physical information, so the target can not be reduced to particle. Thus one node can not get complete target information, and more nodes is needed to work together to cover a target. In this paper, through the targets'virtual gravitation and nodes'virtual attraction and repulsion, the vision sensor nodes can move to the target near and be evenly distributed. Then use particle swarm optimization advantage to adjust node direction and achieve target maximize coverage.