Two‐stage 3D model‐based UAV pose estimation: A comparison of methods for optimization

Abstract

Particle Filters (PFs) have been successfully used in three-dimensional (3D) modelbased pose estimation. Typically, these filters depend on the computation of importance weights that use similarity metrics as a proxy to approximate the likelihood function. In this paper, we explore the use of a two-stage 3D model-based approach based on a PF for single-frame pose estimation. First, we use a classifier trained in a synthetic data set for Unmanned Aerial Vehicle (UAV) detection and a pretrained database indexed by bounding boxes properties to obtain an initial rough pose estimate. Second, we employ optimization algorithms to optimize the used similarity metrics and decrease the obtained error. We have tested four different algorithms: (a) Particle Filter Optimization (PFO), (b) Particle Swarm Optimization (PSO), (c) modified PSO, and (d) an approach based on the evolution strategies present in the genetic algorithms named Genetic Algorithm-based Framework (GAbF). To check the quality of the estimate on each iteration, we have tested several similarity metrics (color, edge, and mask-based) based on the UAV Computer-Aided Design (CAD) model. The framework is applied to the outdoor pose estimation of a fixed-wing UAV for autonomous landing in a Fast Patrol Boat (FPB). We extend our previous approach by adopting a better problem formulation, using Deep Neural Networks (DNNs) for UAV detection, making the comparison between the used similarity metrics, comparing pose optimization schemes, and showing new results.The future work will focus on the inclusion of this scheme in a tracking architecture to increase the accuracy of the result between observations.

Fixed-Wing Unmanned Aerial Vehicle 3D-Model-Based Tracking for Autonomous Landing

“…is the angular velocity according to the camera reference frame represented in Figure 8. As described initially in Section 1, the proposed tracking architecture is divided into (Figure 9) [41][42][43]87]:…”

“…This stage was initially inspired by the Boosted Particle Filter (BPF) [88], which extends the Mixture Particle Filter (MPF) application [89] by incorporating Adaptive Boosting (AdaBoost) [90]. We adopted the approach described in [41][42][43]87], characterized by the following two stages:…”

“…The adopted tracking architecture is divided into: (i) Pose Boosting; (ii) Tracking; (iii) Pose optimization [41,42]. In the Pose Boosting stage, we used Machine Learning (ML) algorithms to detect the UAVs on the captured frame and used a pre-trained database to generate representative pose samples [12,43]. In the Tracking stage, we used a 3D-modelbased tracking approach using the UAV CAD model based on a PF approach [42].…”

“…As far as we know, there are no publicly available datasets or other ground-based model-based UAV tracking approaches, making the comparison with other state-of-the-art methods impossible. The main developments regarding our previous work are: (i) in [42], a tracking architecture using a UBiGaF without pose optimization implementation, which is explored in this article, was proposed; (ii) in [41], we applied optimization steps to the UBiF that is implemented in this article also to the UBiGaF; (iii) in [43], we proposed a distance transform similarity metric that is modified to a tree-based approach to be able to decrease the processing time without a loss in accuracy.…”

Estimation of a Human-Maneuvered Target Incorporating Human Intention

“…Most dynamic targets to track or engage are either human-maneuvered or humans themselves. Estimating the state of such a human-maneuvered target is essential and important, and has attracted tremendous interest in the last decades [ 1 , 2 , 3 , 4 ]. Despite the importance, difficulty in the estimation of the human-maneuvered target lies in the motion uncertainty.…”