Object detection is a fundamental computer vision task for many real-world applications. In the maritime environment, this task is challenging due to varying light, view distances, weather conditions, and sea waves. In addition, light reflection, camera motion and illumination changes may cause to false detections. To address this challenge, we present three fusion architectures to fuse two imaging modalities: visible and infrared. These architectures can provide complementary information from two modalities in different levels: pixel-level, feature-level, and decision-level. They employed deep learning for performing fusion and detection. We investigate the performance of the proposed architectures conducting a real marine image dataset, which is captured by color and infrared cameras on-board a vessel in the Finnish archipelago. The cameras are employed for developing autonomous ships, and collect data in a range of operation and climatic conditions. Experiments show that feature-level fusion architecture outperforms the state-of-the-art other fusion level architectures.

Autonomous harvesting and transportation is a long-term goal of the forest industry. One of the main challenges is the accurate localization of both vehicles and trees in a forest. Forests are unstructured environments where it is difficult to find a group of significant landmarks for current fast feature-based place recognition algorithms. This paper proposes a novel approach where local point clouds are matched to a global tree map using the Delaunay triangularization as the representation format. Instead of point cloud based matching methods, we utilize a topology-based method. First, tree trunk positions are registered at a prior run done by a forest harvester. Second, the resulting map is Delaunay triangularized. Third, a local submap of the autonomous robot is registered, triangularized and matched using triangular similarity maximization to estimate the position of the robot. We test our method on a dataset accumulated from a forestry site at Lieksa, Finland. A total length of 200 m of harvester path was recorded by an industrial harvester with a 3D laser scanner and a geolocation unit fixed to the frame. Our experiments show a 12 cm s.t.d. in the location accuracy and with real-time data processing for speeds not exceeding 0.5 m/s. The accuracy and speed limit are realistic during forest operations.

This work focuses on the development of an effective collision avoidance algorithm that detects and avoids obstacles autonomously in the vicinity of a potential collision by using a single ultrasonic sensor and controlling the movement of the vehicle. The objectives are to minimise the deviation from the vehicle’s original path and also the development of an algorithm utilising one of the cheapest sensors available for very lost cost systems. For instance, in a scenario where the main ranging sensor malfunctions, a backup low cost sensor is required for safe navigation of the vehicle while keeping the deviation to a minimum. The developed algorithm utilises only one ultrasonic sensor and approximates the front shape of the detected object by sweeping the sensor mounted on top of the unmanned vehicle. In this proposed approach, the sensor is rotated for shape approximation and edge detection instead of moving the robot around the encountered obstacle. It has been tested in various indoor situations using different shapes of objects, stationary objects, moving objects, and soft or irregularly shaped objects. The results show that the algorithm provides satisfactory outcomes by entirely avoiding obstacles and rerouting the vehicle with a minimal deviation.

Abnormal behavior detection is currently receiving much attention because of the availability of marine equipment and data allowing maritime agents to track vessels. One of the most popular tools for developing an efficient anomaly detection system is the Automatic Identification System (AIS). The aim of this paper is to explore the performance of existing well-known clustering methods for detecting the two most dangerous abnormal behaviors based on the AIS. The methods include K-means, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), Affinity Propagation (AP), and the Gaussian Mixtures Model (GMM). In order to evaluate the performance of the clustering methods, we also used the AIS data of vessels, which were collected through the Finnish transport agency from the whole Baltic Sea for three months. Although most existing studies focus on ocean route recognition, deviations from regulated ocean routes, or irregular speed, we focused on dark ships or those sets of vessels that turn off the AIS to perform illegal activities and spiral vessel movements. The experimental results demonstrate that the K-means clustering method can effectively detect dark ships and spiral vessel movements, which are the most threatening events for maritime safety.

Two important aspects in dealing with autonomous navigation of a swarm of drones are collision avoidance mechanism and formation control strategy; a possible competition between these two modes of operation may have negative implications for success and efficiency of the mission. This issue is exacerbated in the case of distributed formation control in leader-follower based swarms of drones since nodes concurrently decide and act through individual observation of neighbouring nodes’ states and actions. To dynamically handle this duality of control, a plan of action for multi-priority control is required. In this paper, we propose a method for formation-collision co-awareness by adapting the thin-plate splines algorithm to minimize deformation of the swarm’s formation while avoiding obstacles. Furthermore, we use a non-rigid mapping function to reduce the lag caused by such maneuvers. Simulation results show that the proposed methodology maintains the desired formation very closely in the presence of obstacles, while the response time and overall energy efficiency of the swarm is significantly improved in comparison with the existing methods where collision avoidance and formation control are only loosely coupled. Another important result of using non-rigid mapping is that the slowing down effect of obstacles on the overall speed of the swarm is significantly reduced, making our approach especially suitable for time critical missions.

The rut formation during forest operations is an undesirable phenomenon. A methodology is being proposed to measure the rut depth distribution of a logging site by photogrammetric point clouds produced by unmanned aerial vehicles (UAV). The methodology includes five processing steps that aim at reducing the noise from the surrounding trees and undergrowth for identifying the trails. A canopy height model is produced to focus the point cloud on the open pathway around the forest machine trail. A triangularized ground model is formed by a point cloud filtering method. The ground model is vectorized using the histogram of directed curvatures (HOC) method to produce an overall ground visualization. Finally, a manual selection of the trails leads to an automated rut depth profile analysis. The bivariate correlation (Pearson’s r) between rut depths measured manually and by UAV photogrammetry is $r=0.67$ . The two-class accuracy a of detecting the rut depth exceeding 20 cm is $a=0.65$ . There is potential for enabling automated large-scale evaluation of the forestry areas by using autonomous drones and the process described.