Prediction of Multiple Object Tracking Based on Deep Learning Techniques

Abstract

In recent years, there has been a growing interest in wireless sensor networks (WSNs) because of their potential usage in a wide variety of applications such as remote environmental monitoring and target tracking. Target tracking is an important and common use of wireless sensor networks. Object tracking methods can be classified as continuous monitoring and scheduled monitoring. Energy efficiency and accuracy are the two important issues to be considered in the design of any object tracking algorithm. However, achieving high accuracy of tracking together with energy efficiency in target tracking algorithms is extremely challenging. The traditional target tracking algorithm for multi-target tracking has been numerous, but in the context of WSNs, there are few multi-target tracking schemes Therefore, this thesis is about tracking multiple objects and predicting the movement of these objects in WSNs using deep learning. The aim of this work are increasing tracking accuracy, reducing the energy used in the tracking process, and increasing the network lifetime by using deep learning, so the Long Short Term Memory (LSTM) networks which are the extension of Recurrent Neural Networks (RNNs) is used. Since the tracking process is sequential, the LSTM will be effective in this field. An efficient object tracking system is proposed by combining dynamic and static clustering techniques with the predictive tracking technique using LSTM. Two models of LSTM are used, where the first model classifies the direction of the object movement and calculates the new position of the object. The second model is used to predict the three nodes that track the object, and this system is to track a single object and multiple objects and compute the energy consumption for these two systems, and proposed a new algorithm called Hybrid K_Mean PSO Clustering algorithm (HKM_PSO) to improve the energy consumption and increase the lifetime of the network. Through the results obtained and since the location of the object is random, as well as the distribution of sensors is random, every time the program is executed, the accuracy is very high. Tracking accuracy ranged from 95% to 98% percent, depending on the path length of the object's movement. For tracking a single XI object, the lowest error rate was 0.6, while the greatest error rate was 0.8.for tracking a single object. In addition, to that for tracking multiple objects the tracking accuracy ranged between 94% and 97%, and the lowest error rate was 0.7, demonstrating the accuracy and efficiency of the proposed works. In terms of calculating the consumed energy, the proposed (HKM_PSO) algorithm was very effective in the process of optimizing power and increasing network lifetime