Detection of small moving objects is an important research area with applications including monitoring of flying insects, studying their foraging behavior, using insect pollinators to monitor flowering and pollination of crops, surveillance of honeybee colonies, and tracking movement of honeybees. However, due to the lack of distinctive shape and textural details on small objects, direct application of modern object detection methods based on convolutional neural networks (CNNs) shows considerably lower performance. In this paper we propose a method for the detection of small moving objects in videos recorded using unmanned aerial vehicles equipped with standard video cameras. The main steps of the proposed method are video stabilization, background estimation and subtraction, frame segmentation using a CNN, and thresholding the segmented frame. However, for training a CNN it is required that a large labeled dataset is available. Manual labelling of small moving objects in videos is very difficult and time consuming, and such labeled datasets do not exist at the moment. To circumvent this problem, we propose training a CNN using synthetic videos generated by adding small blob-like objects to video sequences with real-world backgrounds. The experimental results on detection of flying honeybees show that by using a combination of classical computer vision techniques and CNNs, as well as synthetic training sets, the proposed approach overcomes the problems associated with direct application of CNNs to the given problem and achieves an average F1-score of 0.86 in tests on real-world videos.

An accurate and reliable estimation of the transformation matrix between an optical sensor and a robot is a key aspect of the hand–eye system calibration process in vision-guided robotic applications. This paper presents a novel approach to markerless hand–eye calibration that achieves streamlined, flexible, and highly accurate results, even without error compensation. The calibration procedure is mainly based on using the robot’s tool center point (TCP) as the reference point. The TCP coordinate estimation is based on the robot’s flange point cloud, considering its geometrical features. A mathematical model streamlining the conventional marker-based hand–eye calibration is derived. Furthermore, a novel algorithm for the automatic estimation of the flange’s geometric features from its point cloud, based on a 3D circle fitting, the least square method, and a nearest neighbor (NN) approach, is proposed. The accuracy of the proposed algorithm is validated using a calibration setting ring as the ground truth. Furthermore, to establish the minimal required number and configuration of calibration points, the impact of the number and the selection of the unique robot’s flange positions on the calibration accuracy is investigated and validated by real-world experiments. Our experimental findings strongly indicate that our hand–eye system, employing the proposed algorithm, enables the estimation of the transformation between the robot and the 3D scanner with submillimeter accuracy, even when using the minimum of four non-coplanar points for calibration. Our approach improves the calibration accuracy by approximately four times compared to the state of the art, while eliminating the need for error compensation. Moreover, our calibration approach reduces the required number of the robot’s flange positions by approximately 40%, and even more if the calibration procedure utilizes just four properly selected flange positions. The presented findings introduce a more efficient hand–eye calibration procedure, offering a superior simplicity of implementation and increased precision in various robotic applications.