Fusion of Multilayer Perception and Kalman Filter for Indoor Object Tracking

Abstract

Abstract

The task of tracking indoor objects presents a formidable challenge due to a multitude of factors, including but not limited to occlusions, fluctuations in lighting conditions, and intricate object movements. The conventional employment of Kalman filter-based techniques for tracking objects within indoor environments has been extensively utilised. However, these methods frequently encounter constraints such as insufficient flexibility and inadequate depiction of intricate object kinetics. In order to overcome these constraints, a new methodology is suggested which integrates the Kalman filter and Multilayer Perceptron (MLP) models for the purpose of tracking objects within indoor environments. The suggested methodology amalgamates the advantages of the two models, wherein the Kalman filter manages the sensor data that is prone to noise and offers state estimation, whereas the MLP model captures the intricate nonlinear dynamics of the object being tracked. The results obtained from experiments conducted on a dataset that is available to the public demonstrate that the method proposed exhibits superior performance in terms of both tracking accuracy and robustness when compared to existing state-of-the-art methods. The approach being proposed exhibits potential applications in diverse domains including surveillance, robotics, and human-computer interaction, wherein dependable object tracking is of utmost importance.