Performance Comparison of Computationally Efficient Algorithms for Drone Localization on Embedded Systems

Accuracy and computational complexity are the two most fundamental and critical factors for ensuring efficient performance of a real-time drone localization algorithm. This paper explores the time complexity of two popular state estimation algorithms: The Error State Kalman Filter (ESKF), and Particle Filter (PF). The capacity to estimate the state of dynamic systems in the presence of noise makes these algorithms popular in a wide range of applications, including robotics, navigation, and sensor fusion. The main objective of this study is to assess each algorithm’s computational performance, particularly with regard to its time complexity. The paper offers a thorough examination of the variables that affect these algorithms’ computational cost, such as their mathematical structure, the underlying systems’ characteristics, and the complexity of the required operations. Through this investigation, the study hopes to shed light on the trade-offs associated with choosing a suitable state estimation method in terms of computational efficiency. With the simulation results in this work, the computational time obtained with the proposed ESKF algorithm is found to be 0.21 seconds on average and for PF algorithm it is 0.54 milliseconds using the NVIDIA Jetson Xavier hardware platform. The PF’s faster runtime stems from its simplified motion and observation models, which maintain sufficient accuracy under test conditions. This concludes that the proposed PF strategy in this paper is more suitable for real-time drone localization on embedded systems.