ViVa-SAFELAND: a New Freeware for Safe Validation of Vision-based Navigation in Aerial Vehicles

TL;DR

ViVa-SAFELAND addresses the lack of safe, controlled validation tools for vision-based UAV navigation in urban environments by replaying real-world aerial videos with moving obstacles on an Emulated Aerial Vehicle (EAV) equipped with a Virtual Aerial Camera (VAC). It models UAV dynamics with a discrete second-order equation and renders VAC views to enable perception testing, including a Super-Resolution step to preserve detail: $ X_k \approx 2 X_{k-1} - X_{k-2} + (\Delta_t^2/m) ( R_k T_k e_3 - m g e_3 - (k_d/\Delta_t) (X_{k-1} - X_{k-2}) + \xi_k ) $. The paper demonstrates two case studies—semantic segmentation for risk assessment and real-time object detection with YOLOv8—showing safe, repeatable evaluation of landing and navigation strategies under realistic scenes. By offering a fair baseline, dataset-generation capability, and pilot training support, ViVa-SAFELAND facilitates safer, faster development of vision-based UAV solutions for urban deployment.

Abstract

ViVa-SAFELAND is an open source software library, aimed to test and evaluate vision-based navigation strategies for aerial vehicles, with special interest in autonomous landing, while complying with legal regulations and people's safety. It consists of a collection of high definition aerial videos, focusing on real unstructured urban scenarios, recording moving obstacles of interest, such as cars and people. Then, an Emulated Aerial Vehicle (EAV) with a virtual moving camera is implemented in order to ``navigate" inside the video, according to high-order commands. ViVa-SAFELAND provides a new, safe, simple and fair comparison baseline to evaluate and compare different visual navigation solutions under the same conditions, and to randomize variables along several trials. It also facilitates the development of autonomous landing and navigation strategies, as well as the generation of image datasets for different training tasks. Moreover, it is useful for training either human of autonomous pilots using deep learning. The effectiveness of the framework for validating vision algorithms is demonstrated through two case studies, detection of moving objects and risk assessment segmentation. To our knowledge, this is the first safe validation framework of its kind, to test and compare visual navigation solution for aerial vehicles, which is a crucial aspect for urban deployment in complex real scenarios.

Figures (5)

• Figure 1: ViVa-SAFELAND: a Visual Validation Safe Landing tool. An Emulated Aerial Vehicle is implemented to navigate inside aerial videos of real urban environments, from where the drone's view is recovered and used to evaluate vision algorithms.
• Figure 2: Architecture of the network used to apply super-resolution to the images captured by the VAC at low altitudes.
• Figure 3: Comparison of the original images captured by the VAC (top) and the images restored by the super-resolution network (bottom) under different altitudes from 110m to 200m.
• Figure 4: Example of ViVa-SAFELAND operation. The larger left image presents the original video frame, where the VAC RoI is represented by a red box, the EAV velocity vector is represented by the blue arrow for the horizontal coordinates, and as a green circle for the z-axis velocity. The VAC view is depicted in the central right box, where YOLO is implemented to identify critical obstacles such as people and cars. In the upper right box, a U-Net is used for risk assessment using image segmentation. At the bottom right, the drone's flight information is displayed.
• Figure 5: Performance evaluation of U-Net and YOLO outputs for risk assessment and object detection in UAV landing. U-Net (left) shows low-risk areas in black and high-risk zones, like vehicles and people, in red. YOLO (right) detects risk objects with red bounding boxes.