In this study, we proposed a system to quantify the plant height from a three-dimensional point cloud of a plant using the INTEL REALSENSE DEPTH CAMERA D415. This system was used to compare and evaluate plant growth under LEP, LED, and fluorescent lighting. In the experiment, four basil plants were hydroponically cultivated as measurement targets. The highest point from the three-dimensional point cloud of each plant reconstructed by D415 was calculated as the plant height, and the change for about one month was recorded. As a result of the experiment, it was confirmed that the plants grow faster in the order of LEP, LED, and fluorescent lamp. Regarding the illuminance, LEP, fluorescent lamp, and LED were the highest in this order. LEP is advantageous for plant growth because it can emit light with a wide band spectrum with high efficiency. Cultivation using LED tended to grow faster than fluorescent lamps. Since the LED emits only light in the wavelength band related to plant growth, it was considered that efficient plant growth was realized. It was confirmed that the proposed measurement system not only enables easy quantification of plant height, but also enables visualization of growth by presenting a three-dimensional point cloud. Under artificial light illumination, it became clear that the accuracy of 3D reconstruction differs between when artificial light is on and when it is off. Under artificial light lighting, the reconstruction accuracy tends to be lower, and improvement is a future task.
In this paper, we propose an animal blink detection method using the frame subtraction method for the purpose of automating physical examination by measuring the blink frequency in a veterinary hospital. The proposed method is based on an algorithm that focuses on the features of the frame subtraction images and is robust to the luminance changes caused by the animal’s body movements. In addition, the proposed system does not require any pre-learning and runs in real-time. For evaluation, we conducted a blink detection experiment using free videos of birds available to the public. Experimental results suggest that the proposed method can be used as a practical blink detector for animals.
In this paper, we design both an individual blade pitch angle controller and a collective blade pitch angle controller using wind speed preview information for floating offshore wind turbines. The controllers are designed to reduce fluctuations of the rotor speed, platform motions, and load on blades. We apply H2 preview control theory to the linearized model of the floating offshore wind turbine for designing the controllers. We compare the individual blade pitch angle controller with the collective blade pitch angle controller using the high-fidelity wind turbine simulator, FAST. Simulation results show that the individual blade pitch angle controller using incoming wind speed preview suppresses the fluctuations of the rotor speed, platform pitch rate, and load on blades.
In this paper, we investigate an assist control for a system in which a human and a robot cooperate to transport an object toward a designated position and angle. We consider a situation where the force due to the human to the object cannot be measured directly. In order to tackle this situation, an assist control with a nonlinear disturbance observer is proposed. By using the observer, we estimate the control input due to the human from position and angle of the object. Moreover, we design the control input by the robot to assist the human input based on the estimates by the disturbance observer. The effectiveness of the proposed method is discussed by numerical simulations for the minimum jerk model.
In this paper, we propose an efficient method for human dense avoidance based on a coverage control. Our motivation is to avoid crowding in public facilities such as stations and government offices, and human dense in the current situation of COVID-19 from system and control theory. In this paper, humans and robots are modelled as heterogeneous and homogeneous agents, respectively, which make decisions based on their local information. We suppose a dense situation caused by the rendezvous among humans due to their own inherent dynamics. As a main result, we propose a coverage control for a distributed movement of multiple humans. We also characterize the stationary point analytically in terms of the the gains which describe a strength of the interconnection of the agents, and the centers of the Voronoi regions related to the agents. Moreover, we verify the meaning of the characterization from an engineering viewpoint of the dense avoidance. Finally, we show the efficiency of the method based on a numerical simulation.
The purpose of this study is to construct an automatic control form that can stably and precisely control the movement of a balloon robot by applying a control system using a temporal state control form with the movement distance as the time axis. Using this system, simulations were conducted for multiple paths to confirm the performance of the control system. As a result of the simulation, it was confirmed that the arrival speed to the target axis and the trajectory can be changed by changing the newly introduced gain. In addition, the same control system was able to track a complex path combining straight lines in the simulation. These results show that the control system developed in this study can be controlled more precisely than the time-state control form with the x-axis as the time axis, and also has high versatility by adjusting the gain.
A recommendation system is one of the methods to support users' information acquisition. Collaborative filtering (CF) is a popular method to achieve it. Users' preferences (how much users like items) are represented as rating matrix for using CF. However, the percentage of missing ratings (called sparsity) is usually high, which results in poor performance. To address this problem, a hybrid method has been proposed to mitigate the performance degradation even at high sparsity by combining not only the ratings but also other information about items and users. Collaborative topic regression (CTR) is a pioneer model, which extracts latent topics from documents about items or users and uses them together with rating matrix. However, due to the increase of model complexity of CTR compared to CF, the expected performance will not be achieved unless the word selection in the document and the hyperparameter settings should be done appropriately. In this study, we experimentally show how the performance of CTR changes under different sparsity levels, by varying the words selected based on document frequency and by varying the combination of hyperparameters that adjust the influence of the ratings and that of the latent topics. We prepare three datasets with different domains and evaluate the generality of the results in our experiments.