Image-based refined 3D reconstruction relies on high-resolution and multi-angle images of scenes. The assistance of multi-rotor drones and gimbal provides great convenience for image acquisition. However, capturing images with manual control generally takes a long time. It could easily lead to redundant or insufficient local area coverage, resulting in poor quality of the reconstructed model. We propose a surface geometric primitive-guided UAV path planning method (SGP-G) that aims to automatically and quickly plan a collision-free path to capture fewer images, based on which high-quality models can be obtained. The geometric primitives are extracted by plane segmentation on the proxy, which performs three main functions. First, a more representative evaluation of the reconstructability of the whole scene is realized. Second, two optimization strategies for different geometric primitives are executed to quickly generate a near-global optimized set of viewpoints. Third, regularly arranged viewpoints are generated to improve the efficiency of image acquisition. Experiments on virtual and real scenes demonstrate the remarkable performance of our method. Compared with the state of the art, we accomplish the planning of the photographic path with higher efficiency in a relatively simple way, achieving equivalent and even higher quality of the reconstructed model with fewer images.

Path planning and obstacle avoidance are pivotal for intelligent unmanned aerial vehicle (UAV) systems in various domains, such as postdisaster rescue, target detection, and wildlife conservation. Currently, reinforcement learning (RL) has become increasingly popular in UAV decision-making. However, the RL approaches confront the challenges of partial observation and large state space when searching for random targets through continuous actions. This paper proposes a representation enhancement-based proximal policy optimization (RE-PPO) framework to address these issues. The representation enhancement (RE) module consists of observation memory improvement (OMI) and dynamic relative position-attitude reshaping (DRPAR). OMI reduces collision under partially observable conditions by separately extracting perception features and state features through an embedding network and feeding the extracted features to a gated recurrent unit (GRU) to enhance observation memory. DRPAR compresses the state space when modeling continuous actions by transforming movement trajectories of different episodes from an absolute coordinate system into different local coordinate systems to utilize similarity. In addition, three step-wise reward functions are formulated to avoid sparsity and facilitate model convergence. We evaluate the proposed method in three 3D scenarios to demonstrate its effectiveness. Compared to other methods, our method achieves a faster convergence during training and demonstrates a higher success rate and a lower rate of timeout and collision during inference. Our method can significantly enhance the autonomy and intelligence of UAV systems under partially observable conditions and provide a reasonable solution for UAV decision-making under uncertainties.

The shadow-mapping and ray-tracing algorithms are the two popular approaches used in visibility handling for multi-view based texture reconstruction. Visibility testing based on the two algorithms needs a user-defined bias to reduce computation error. However, a constant bias does not work for every part of a geometry. Therefore, the accuracy of the two algorithms is limited. In this paper, we propose a high-precision graphics pipeline-based visibility classification (GPVC) method without introducing a bias. The method consists of two stages. In the first stage, a shader-based rendering is designed in the fixed graphics pipeline to generate initial visibility maps (IVMs). In the second stage, two algorithms, namely, lazy-projection coverage correction (LPCC) and hierarchical iterative vertex-edge-region sampling (HIVERS), are proposed to classify visible primitives into fully visible or partially visible primitives. The proposed method can be easily implemented in the graphics pipeline to achieve parallel acceleration. With respect to efficiency, the proposed method outperforms the bias-based methods. With respect to accuracy, the proposed method can theoretically reach a value of 100%. Compared with available libraries and software, the textured model based on our method is smoother with less distortion and dislocation.