Improving Node Localization Accuracy in Wireless Sensor Networks based on Computer Vision and Deep Learning Optimization

Abstract

In order to solve the problem of angular effects and reduced positioning accuracy caused by rapid speed changes in position tracking and positioning methods in wireless sensor networks, as well as the difficulty of improving positioning accuracy with a single solution, the author proposes a research on improving node positioning accuracy in wireless sensor networks based on computer vision and deep learning optimization. The author proposes a tracking and localization method using Kalman filtering (KF) and visual assistance on the TI CC2431 ZPS platform. On the basis of normalized cross-correlation, visual assistance calibration technology is used to extract the position of reference nodes as landmarks using visual assistance methods. Then, the KF method is used to calibrate the position estimation, which randomly generates virtual nodes for neural network training. Then, the priority positioning node is located and used as the anchor node for the next positioning, and the wireless loop is used for positioning calculation. The experimental results show that both the TI ZPS method and the KF based method have an estimated position error distance of over 55%, which is less than 2.2m and 1.8m, respectively, The proposed tracking and positioning method has an estimated position error distance of over 55% less than 1.4m. The method proposed by the author effectively avoids uncertainty caused by system errors in actual dynamic environments, reduces angular effects in position estimation systems, and improves positioning accuracy.