Search

The use of drones or Unmanned Aerial Vehicles (UAVs) and other flying vehicles has increased exponentially in the last decade. These devices pose a serious threat to helicopter pilots who constantly seek to maintain situational awareness while flying to avoid objects that might lead to a collision. In this paper, an Airborne Visual Artificial Intelligence System is proposed that seeks to improve helicopter pilots’ situational awareness (SA) under UAV-congested environments. Specifically, the system is capable of detecting UAVs, estimating their distance, predicting the probability of collision, and sending an alert to the pilot accordingly. To this end, we aim to combine the strengths of both spatial and temporal deep learning models and classic computer stereo vision to (1) estimate the depth of UAVs, (2) predict potential collisions with other UAVs in the sky, and (3) provide alerts for the pilot with regards to the drone that is likely to collide. The feasibility of integrating artificial intelligence into a comprehensive SA system is herein illustrated and can potentially contribute to the future of autonomous aircraft applications.

Pan–tilt–zoom cameras are commonly used for surveillance applications. Their automation could reduce the workload of human operators and increase the safety of airports by tracking anomalous objects such as drones. Reinforcement learning is an artificial intelligence method that outperforms humans on certain specific tasks. However, there exists a lack of data and benchmarks for pan–tilt–zoom control mechanisms in tracking airborne objects. Here, we show a simulated environment that contains a pan–tilt–zoom camera being used to train and evaluate a reinforcement learning agent. We found that the agent can learn to track the drone in our basic tracking scenario, outperforming a solved scenario benchmark value. The agent is also tested on more complex scenarios, where the drone is occluded behind obstacles. While the agent does not quantitatively outperform the optimal human model, it shows qualitative signs of learning to solve the complex, occluded non-linear trajectory scenario. Given further training, investigation, and different algorithms, we believe a reinforcement learning agent could be used to solve such scenarios consistently. Our results demonstrate how complex drone surveillance tracking scenarios may be solved and fully autonomized by reinforcement learning agents. We hope our environment becomes a starting point for more sophisticated autonomy in control of pan–tilt–zoom cameras tracking of drones and surveilling airspace for anomalous objects. For example, distributed, multi-agent systems of pan–tilt–zoom cameras combined with other sensors could lead towards fully autonomous surveillance, challenging experienced human operators.

We present a convolutional neural network (CNN) that identifies drone models in real-life videos. The neural network is trained on synthetic images and tested on a real-life dataset of drone videos. To create the training and validation datasets, we show a method of generating synthetic drone images. Domain randomization is used to vary the simulation parameters such as model textures, background images, and orientation. Three common drone models are classified: DJI Phantom, DJI Mavic, and DJI Inspire. To test the performance of the neural network model, Anti-UAV, a real-life dataset of flying drones is used. The proposed method reduces the time-cost associated with manually labelling drones, and we prove that it is transferable to real-life videos. The CNN achieves an overall accuracy of 92.4%, a precision of 88.8%, a recall of 88.6%, and an f1 score of 88.7% when tested on the real-life dataset.

Visual navigation is an essential part of planetary rover autonomy. Rock segmentation emerged as an important interdisciplinary topic among image processing, robotics, and mathematical modeling. Rock segmentation is a challenging topic for rover autonomy because of the high computational consumption, real-time requirement, and annotation difficulty. This research proposes a rock segmentation framework and a rock segmentation network (NI-U-Net++) to aid with the visual navigation of rovers. The framework consists of two stages: the pre-training process and the transfer-training process. The pre-training process applies the synthetic algorithm to generate the synthetic images; then, it uses the generated images to pre-train NI-U-Net++. The synthetic algorithm increases the size of the image dataset and provides pixel-level masks—both of which are challenges with machine learning tasks. The pre-training process accomplishes the state-of-the-art compared with the related studies, which achieved an accuracy, intersection over union (IoU), Dice score, and root mean squared error (RMSE) of 99.41%, 0.8991, 0.9459, and 0.0775, respectively. The transfer-training process fine-tunes the pre-trained NI-U-Net++ using the real-life images, which achieved an accuracy, IoU, Dice score, and RMSE of 99.58%, 0.7476, 0.8556, and 0.0557, respectively. Finally, the transfer-trained NI-U-Net++ is integrated into a planetary rover navigation vision and achieves a real-time performance of 32.57 frames per second (or the inference time is 0.0307 s per frame). The framework only manually annotates about 8% (183 images) of the 2250 images in the navigation vision, which is a labor-saving solution for rock segmentation tasks. The proposed rock segmentation framework and NI-U-Net++ improve the performance of the state-of-the-art models. The synthetic algorithm improves the process of creating valid data for the challenge of rock segmentation. All source codes, datasets, and trained models of this research are openly available in Cranfield Online Research Data (CORD).