Remote Sensing for Agricultural Crops Based on a Low Cost Quadcopter

This paper presents a proposal for information gathering from crops by means of a low-cost quadcopter known as the AR Drone 2.0. To achieve this, we designed a system for remote sensing that addresses challenges identified in the present research, such as acquisition of aerial photographs of an entire crop and AR Drone navigation on non-planar areas arises. The project is currently at an early stage of development. The first stage describes platform and hardware/software tools used to build the proposed prototype. Second stage characterizes performance experiments of sensors stability and altitude in AR Drone, in order to design an altitude strategy control over non-flat crops. In addition, path planning algorithms based on shortest route by graphs (Dijkstra, A* and wavefront propagation) are evaluated with simulated quadcopter. The implementation of the shortest path algorithms is the beginning to full coverage of a crop. Observations of quadcopter behavior in Gazebo simulator and real tests demonstrate viability to execute the project by using AR Drone like platform of a remote sensing system to precision agriculture.


I. Introduction
Unmanned Aerial Vehicles (UAV) offer serious opportunities in the design of systems intended to measure and obtain physical and chemical properties of several phenomena.This new technological approach is the result of platforms characteristics, with multiple sensors, capable of achieving object features with a higher level of detail.The reach of these UAV is related to remote sensing, which is a method of obtaining information without direct contact by using devices such as cameras, lasers, radiofrequency (RF) receivers, radars, sonars, and magnetometers.Precision agriculture is one of the remote sensing related areas, since agricultural labor requires acquisition, processing, and interpretation of higher data volumes coming from broad areas in a very precise way (Joseph, 2005).Precision agriculture relies on remote sensing to manage resources efficiently in adequate place, time, and moment; avoiding economic loss (Bongiovanni & Lowenberg-Deboer, 2004).Remote sensing allows data gathering of temporal and spatial variability of plants, soil, and environment parameters.These variability parameters are obtained from photograph mosaics by analyzing the electromagnetic spectrum with the aim of obtaining vegetation indices and, consequently, determining real condition of crops.In consequence, this leads to the establishment of damage and/or evolution in crop periods (i.e.sowing, growing, maturing, and harvesting, among others), and medium range diagnoses (Ji-hua & Bing-fang, 2008).
Under the precision agriculture domain, remote sensing systems that include several UAV types are used.Most of these UAVs weigh between 1 and 20 kilograms and reach altitudes up to 900 meters, depending on their payload.Currently, trends using these devices follow features previously presented, due to their ease of modeling and operation together with their low cost.To illustrate some examples, Carolo P330 UAV was one of the first aerial devices used in crop photogrammetry; it is a fixed wing device with manual control and GPS-based autopilot (Grenzdorffer, Engel, & Teichert, 2008).Helicopter-shaped models are also used in agricultural tasks, managing to capture georeferenced images using GPS and specialized payload such as the Febry-Perot interferometer (Sugiura, Noguchi, & Ishii, 2005).However, robots with multiple rotors (i.e.quadcopters, hexacopters, octocopters, etc.) are the most popular UAVs in agricultural labor and their use to obtain vegetation indices and crop mapping is increasing.

I. Introducción
Las aeronaves no tripuladas conocidas como VANT [vehículos aéreos no tripulados] ofrecen grandes oportunidades en el diseño de sistemas para la medición y el registro de propiedades físicas y químicas de fenómenos u objetos en áreas extensas.Este nuevo enfoque tecnológico se debe a que las plataformas están dotadas con múltiples sensores, mediante los cuales es posible conseguir características de los objetos con mayor detalle.El alcance de estos vehículos aéreos está reflejado en la teledetección, entendiéndola como un método para obtener información de las propiedades sin contacto directo, con dispositivos tales como cámaras, láser, receptores de radiofrecuencia, radares, sonares y magnetómetros.La agricultura de precisión es uno de los campos de acción de la teledetección, ya que las prácticas agrícolas requieren la adquisición, el procesamiento y la interpretación de una gran cantidad de datos, de manera precisa, de extensas áreas (Joseph, 2005).La agricultura de precisión se apoya en la teledetección para gestionar eficientemente los recursos en el lugar, tiempo y momento adecuados, evitando pérdidas económicas y productivas (Bongiovanni & Lowenberg-Deboer, 2004).La teledetección permite la adquisición de datos de la variabilidad temporal y espacial de parámetros de plantas, suelo y ambiente.Dicha variabilidad se puede obtener mediante mosaicos de fotografías en los cuales se analiza el espectro electromagnético, para obtener los índices de vegetación y determinar las condiciones reales de cultivos.De esta forma se establece el deterioro y/o evolución en periodos de tiempo, la etapa de crecimiento (siembra, maduración, cosecha, entre otras) y un diagnóstico de mediano alcance Remote Sensing for Agricultural Crops Based on a Low Cost Quadcopter.Sistemas & Telemática,13(34), 49-63 Turner, Lucieer, and Watson (2011) describe a precision viticulture system that uses an octocopter with a Canon® digital camera, a Tetracam® multi-camera array with filters in six bands, and a thermal infrared camera.The system was deployed for 5 minutes in Richmond (Tasmania, Australia) following a cell-shaped trajectory with 200x100 meters resolution.The research project evaluates several algorithms capable of processing ortho-mosaics in each spectrum band.Primicerio et al. (2012) use a remote sensing system designed for vineyards in Italy and based on a micro-copter called VIPtero.The aerial robot is remotely operated for 10 minutes, carrying a multispectral camera to get several vegetation indices.Another research project with a six-rotor configuration, but in color spectrum bands (RGB), is the FlightCopter system.This UAV flies over oats and peas fields located in Hesse, Germany, flying at an elevation of 30 meters and approximately for 10 minutes.With this configuration, authors were able to calculate a normalized differential index to connect ground biomass with the leaf area index in the whole farm (Jannoura, Brinkmann, Uteau, Bruns, & Joergensen, 2015).A representative work in remote sensing systems using quadcopters employs MD4-1000 UAV; it acquires ortho-mosaics from wheat located in Seville (Spain), differentiating weeds from crops (Gómez-Candón, De Castro, & López-Granados, 2014).
For the present work, we selected the AR Drone as aerial platform given its low cost, sensors and remarkable stability (i.e. it uses vision techniques to estimate positioning).The AR Drone is a quadcopter manufactured by Parrot® who have had experience in augmented reality for over 5 years.The AR Drone 2.0 has a control based on speed variation of four rotors added into a cross-shaped symmetric structure and it features several sensors such as the accelerometer, gyroscope, sonar, pressure sensor, and cameras (low and frontal in high definition).Similar to other UAVs, cameras in the AR Drone are physically modeled by six degrees of freedom: "x", "y", "z", "roll", "pitch", and "yaw", where roll, pitch, and yaw are turns in the "x", "y", and "z" axes respectively.These parameters set relative positioning, speed, and orientation of aerial robot.The AR Drone 2.0 has a top-flight time of 15 minutes and a maximum payload of 100 grams; in addition, operation through Wi-Fi from smartphones or laptops is possible (Bristeau, Callou, Vissiere, & Petit, 2011).The applications of the AR Drone seis bandas y una cámara térmica infrarroja.El sistema se despliega durante cinco minutos en Richmond al sur de Tasmania, siguiendo una trayectoria sobre celdas con resolución de 200x100 metros.El trabajo realiza una evaluación de algoritmos para el procesamiento de orto-mosaicos en cada espectro.Primicerio et al. (2012) despliegan un sistema para teledetección de un viñedo en el centro de Italia, con un Mikropkopter de seis rotores, denominado VIPtero.El robot aéreo es tele-operado por diez minutos, portando una cámara multi-espectral, para la obtención de una gran cantidad de índices de vegetación.Otro trabajo con la configuración de seis rotores, pero en el espectro de color (RGB), es el sistema FlightCopter.Este vehículo aéreo es operado sobre cultivos de avena y arveja ubicados en una granja al norte de Hesse en Alemania, con alturas de vuelo de 30 metros, en aproximadamente 10 minutos.Con el sistema se logra calcular el índice diferencial normalizado para relacionar la biomasa sobre el suelo y el índice del área de la hoja en toda la granja (Jannoura, Brinkmann, Uteau, Bruns, & Joergensen, 2015).Un trabajo representativo de teledetección con imágenes digitales capturadas con cuadricópteros utiliza el robot MD4-1000, con el cual obtiene orto mosaicos de cultivos de trigo ubicados en Sevilla (España), para diferenciar la maleza (Gómez-Candón, De Castro, & López-Granados, 2014).
(and UAV in general) are not only for entertainment, but also for health purposes, civil constructions, vehicle surveillance, domiciliary services, and so on.Because of its portability and free development license, the AR Drone allows the convergence of several technologies like Kinect, GPS, and others presented in smartphones, helping with various common activities.In the agricultural field, we could not find related work using this Parrot® technology and device in lands and fields.This is because the construction of the AR Drone is not suitable for wild and rough environments (i.e. its performance might be seriously affected); furthermore, its size and weigh prevent it from being controlled at long distances without damage concerns.Conversely, the AR Drone is ideal to build a remote sensing prototype capable of verifying the design of the proposed system in this document.
As mentioned previously, the general purpose of this project is building a low-cost remote sensing system capable of mapping crops and cultivation precisely.With the purpose of achieving this goal, aerial robot should sample desired crops accurately using aerial photography.A route planner is required to gather complete and precise data from crops.In consequence, software platforms to manage the navigation of UAVs through GPS are essential in this project.These platforms have modules, which select points on a map (usually Google® Maps) and send desired route to the UAV.For instance, QGroundControl is a control station software that connects various autopilots of several UAV types thanks to the MAVLink protocol (Qgroundcontrol.org, 2009).Mission Planner and APM Planner are two software suites that are capable of controlling UAV flight and actual state remotely, as long as the autopilot type is ArdupilotMega or Pixhawk (Open Source Autopilot, 2015).Paparazzy is an open source project known for its control over the AR Drone and Bebop UAV, besides autopilots based on this project.Additionally, there are proper optimized platforms based on application features and developers interests.An interesting case is Naza-M: the main ground control application for the DJI® Phantom quadcopter.In the agricultural field, there is a useful software suite capable of capturing spatial monitoring and diagnosis (Agribotix, 2015).Route planners are also part of the robot coordination tendencies for UAV use in remote sensing systems.Joao Valente details a three-quadcopter system in a farm located in Spain; platforms used in this study were AR 100 and Ast Tec Hummingbird (Valente, 2011).
The rest of this document is structured as follows: section II describes the methodology for system implementation, section III details results of experimental tests and section IV presents conclusions of presented proposal.

II. Methodology
Based on the previous introductory section and references consulted over development of this work, we identified some particular challenges in development of these kinds of systems for agriculture.The first is related with limited payload of UAV; since carrying additional sensors implies lower performance (hovering and maneuverability are affected with bigger payloads).The second issue is low flying autonomy (i.e. the AR Drone provides up to 15 minutes).Another difficult topic is robot configuration, as they usually consider optimal environmental conditions (like low wind speed, flat-shaped crops, and sunny days) for flights.Finally, the last challenge is the limited range of commercial route planners as a result of their manual operation; therefore, area, time, and number of pass points are not specified.In the following subsections, we describe the planned stages to achieve successful development of the proposal.

A. Problem formulation
Remote sensing with UAVs surpasses restrictions presented in traditional sensing in topics related with costs and multiple sensors handling; this allows gathering of detailed information related with physical chemistry variables at any time and place.Nevertheless, sensing based on aerial vehicles is limited by flying time and information losses given system failures.Hence, our approach is oriented to achieving implementation of a low-cost system to monitor spatial and temporal variability of crops through images ( Figure 1).The system must obtain trustworthy information from the studied crop, so, it is necessary to consider features like geography and the flight time of the AR Drone.
To address these problems, we pursue three particular objectives: adjusting the platform to crop features, i.e. optimizing the AR Drone positioning in open environments; obtaining complete crop information through route planning considering quadcopter limitations; and optimizing the system in order to gather useful images while bearing in mind previous items.
Remote Sensing for Agricultural Crops Based on a Low Cost Quadcopter.Sistemas & Telemática,13(34),[49][50][51][52][53][54][55][56][57][58][59][60][61][62][63] the conclusions presented by Vijay Kumar, one of the first promoters of research with quadcopters.Kumar states that positioning estimation and task planning are the biggest challenges of applications with quadcopters.This is since Euler's aerodynamic model (which considers six degrees of freedom) is an approximation that does not involve interaction between engines and external forces; producing uncertainties in controllers and algorithms designed for missions planning.As a result of that, we present an architecture consisting of three development layers, which includes reactive and deliberative behaviors of UAVs to obtain optimal navigation.The design presented in Figure 2 is susceptible to possible changes inherent to implementation; it is also structured considering hybrid control architecture for mobile robots (Arkin, 1990).

Reactive layer
This layer faces challenges related with robot positioning in open fields and its adaptation to agricultural crops.To face this, we evaluate the main positioning techniques of quadcopters.Relative positioning is based on linear controllers for inertial movement and altitude, i.e.Proportional Integral Derivative (PID) (Li & Li, 2011), Linear-Quadratic Regulator (LQR) (Guclu & Arikan, 2012) or combinations of PID and complex controllers, e.g. the Extended Kalman Filter (EKF) (Tanveer, Hazry, Warsi, & Joyo, 2013).Limitations with these approaches rely on positioning based only on inertial movement unit is not functional in open environments.For positioning in external areas, GPS receivers should integrate with the inertial movement unit, achieving geographic location.Nonetheless, the position estimate depends on signal quality and it is not recommendable if high accuracy is needed (Tailanian, Paternain, Rosa, & Canetti, 2014).Another option to manage the AR Drone positioning is by using simultaneous location and mapping techniques with a frontal camera.Still, the high computational complexity needed restricts its adoption for the presented prototype.
As noted before, accuracy search in external positioning concludes in a strategy that combines altitude sensors with GPS receivers.Moreover, even if this approach entails errors due to atmospheric conditions, academic literature shows that it is possible to obtain errors as low Los objetivos planteados por la propuesta están en línea con las conclusiones a las que ha llegado Vijay Kumar, uno de los famosos promotores de investigación con cuadricópteros.Kumar declara que la estimación del posicionamiento y la planificación de tareas son los mayores retos de las aplicaciones futuras con cuadricópteros, ya que el modelo aerodinámico de Euler, que considera 6 grados de libertad, es una aproximación que no involucra la interacción entre los motores y las fuerzas externas ejecutadas sobre los propulsores; generando incertidumbre en los controladores y en el diseño de los algoritmos para la planificación de misión.Desde el anterior escenario se presenta una arquitectura con tres capas de desarrollo que integra comportamientos reactivos y deliberativos del robot aéreo para una óptima navegación.El diseño de la Figura 2 es susceptible a posibles cambios inherentes a la implementación y está estructurado considerando una arquitectura de control hibrida para robots móviles (Arkin, 1990).
Capa reactiva Esta capa enfrenta el reto del posicionamiento del robot en espacios abiertos y la adaptación a los cultivos agrícolas.Para abordarlo son evaluadas las principales técnicas de posicionamiento de un cuadricóptero.La primera de ellas es el posicionamiento relativo, el cual está basado en controladores lineales para los sensores de movimiento inercial y de altura como PID (Proportional-Integral-Deribative) (Li & Li, 2011), LQR (Linear-Quadratic Regulator) (Guclu & Arikan, 2012) o combinaciones de PID con controladores complejos, como por ejemplo, el Filtro de Kalman Extendido (EKF) (Tanveer, Hazry, Warsi, & Joyo, 2013).La limitación con esta iniciativa es que el posicionamiento basado solo en la unidad de movimiento inercial no es funcional en entornos abiertos.Para posicionamiento en exteriores pueden ser integrados receptores GPS (Global Position System) a la unidad de movimiento inercial, logrando una localización geográfica.Sin embargo, la estimación de la posición depende de la calidad de la señal y no es recomendable si la aplicación requiere precisión en la altura ( as 1 meter in altitude and accuracy up to 90% in longitude and latitude (Chee & Zhong, 2013).With precise positioning of the AR Drone over crops, the next step consists of adopting a quadcopter for topography.For this purpose, we focus on non-linear control techniques based on fuzzy logic, since this tactic models the real world in a qualitative manner and it provides certain answers, just as human beings do.Furthermore, implementation of this technique is highly supported by ROS operating system (Bayar, Akar, Yayan, Yavuz, & Yazici, 2014).
Fuzzy control starts with data reception from sensors in a plant (platform), i.e. interpretation of the real world.Data is transformed in fuzzy variables in a process called fuzzification, where it describes plant features (e.g.too high, high, too low plants, etc.).Fuzzy data now reaches a set of behavior rules, which create an inference engine.This engine allows data to be obtained regarding of relations between fuzzy variables and expected results; emerging in fuzzy outputs, which are converted in real values in defuzzification.Hence, fuzzy control permits plants modeling with platform intuitive knowledge and qualitative description of their behavior.For the particular case of altitude control, we found projects results for submarine vehicles, UAV, and zeppelin type vehicles; these research projects demonstrate efficiency of fuzzy control technique and its adaptability to operate with PID combinations and neural networks, among other (Jian-Guo & Jun, 2008) (Shengyi, Kunqin, & Jiao, 2009) (Mehranpour, Emamgholi, Shahri, & Farrokhi, 2013).
Implementation of this reactive layer requires complementary processing between platform and sensors, being capable to control altitude and speed together with Fli-ghtRecorder data (GPS module of AR Drone).

Deliberative layer
This layer deals with objective related with obtaining of information considering platform limitations and coverage area.Expected results of this stage consist on efficient AR Drone navigation in geographic points over crops, avoiding potential damages and information loss.For obtaining these points, we pretend to design a coverage route planner.Regarding consulted literature, there is a mobile robot navigation based on sensors called local planning, and a global planning that requires a limited area with known obstacles for navigation (Galceran & Carreras, 2013).Based on this information and due to project objectives, we choose to design a global planner, since sensors usage for crop explorations implies higher ción para el posicionamiento del AR Drone es usar técnicas de localización y mapeo simultaneo con la cámara frontal.No obstante, la alta complejidad computacional de esta alternativa, limita su adopción para el prototipo propuesto.
La ejecución de esta capa reactiva requiere de un procesamiento complementario entre la plataforma y los sensores, que controle la altura y la velocidad, en conjunto con los datos del FlightRecorder, modulo GPS del AR Drone 2.0.
Even though global planners do not consider navigation in dynamic environments, they beforehand manage turns and longitude presented in routes.In order to achieve this management, general procedure points out in dividing coverage area in grids (i.e.approximate decomposition) or dividing it by exact decomposition (Pignon & Choset, 1998).After that, points where UAVs pass are marked and, subsequently, they determine the coverage plan considering existing restrictions.Some authors consider strategies like the A* algorithm, wavefront based algorithm, and expansion tree to find better coverage routes over crops.This layer evaluates at least two of these algorithms to define best coverage route depending on crop conditions.Implementation of this planner requires relationships between route points (given by route planner) with camera resolution present in UAV and selected flight altitude.

Management layer
The management layer is in charge of implementation of previous tasks and platform control at low and high levels.A low level implies signals captured in the local world, whilst a high level implies signals from worldwide.Low level management infers automatic aerial photography capture in referenced points, whereas high level assumes fulfillment of tasks in the appropriate time to maintain the AR Drone positioning and pre-established route.

A. Development tool
The proposal presented in this document is now at an initial stage, we are updating tools such as the following, in order to progress in desired activities: • Raspberry Pi B+ Model, which is the optional processing unit to handle data from altitude and GPS sensors.We chose this Small Board Computer (SBC) due to its low weight, broad documentation, and acceptable processing capacity.
• Robot Operating System (ROS), basis of coverage planner development, driver implementation, and connection with the AR Drone.We use Fuerte/Hydro ROS over Ubuntu 12.04.
• Gazebo and RVIZ visualization tool, free licensed software suites used to support simulation and validation of navigation and height control strategies.
Capa de gestión Esta capa ejecuta las tareas anteriores y controla las acciones de la plataforma a los niveles bajo y alto, considerando nivel bajo s las señales capturadas del mundo local y nivel alto al modelo del mundo global.La gestión a nivel bajo incluye la captura automática de fotografías aéreas en los puntos referenciados; a nivel alto asume la ejecución de tareas en el tiempo justo para mantener un posicionamiento del AR Drone y el seguimiento de una ruta prestablecida

A. Herramientas de desarrollo
El sistema propuesto en el presente documento se encuentra en una etapa inicial en la cual se están actualizando herramientas para progresar en las actividades formuladas como: • Raspberry Pi Modelo B+, unidad de procesamiento opcional para integrar los datos del sensor de altura y el GPS; los motivos que llevaron a la selección de esta SBC [Small Board Computer] son el peso liviano, la amplia documentación y la significativa capacidad de procesamiento.• El sistema operativo para robots ROS (Robot Operative System), base para el desarrollo del planificador de cobertura, la implementación de controladores y la conexión con el AR Drone; para la implementación actual se utiliza ROS Fuerte/Hydro sobre Ubuntu 12.04.• Gazebo y visualizador RVIZ, herramientas soportadas con licencia libre en cualquier versión de ROS, mediante la que se apoya la simulación y validación de las estrategias de navegación y control de altura.

B. Platform validation and height control
The AR Drone quadcopter consists of a navigation system formed by a three-axis gyroscope, accelerometer, and magnetometer; besides of a pressure sensor, ultrasound sensor (for altitude data), and a QVGA camera (running at 60 FPS for speed measurements).This inertial movement technology allows better piloting maneuvers; besides flight control with absolute reference and wind dynamic estimation (through a magnetometer).In the following figures, we show a comparison between AR Drone behavior in simulation and real environments.Figure 3 shows the speed over the "x" axis (in red), in "y" axis (in blue), and in "z" axis (in green) over a short straight-line trajectory.In the real environment, it is noticeable that external forces act on the quadcopter in approximately 25% of relative error.In contrast, Figure 4 presents results of about 10% of relative error in a simulated scenario.
Because of the previous results, we were forced to enhance the navigation accuracy of the aerial platform.Hence, we selected an integration of geographic positioning over GPS receivers to achieve localization and altitude control through pressure sensors.We discard the
In order to develop previous considerations successfully, the main reference is the research work made by (Mehranpour, Emamgholi, Shahri, & Farrokhi, 2013).Contrary to the proposal presented by these authors, which basically consists in combining fuzzy control with PID control (fuzzy PID) over a quadcopter, our prototype relies control directly in inertial movement unit (i.e. in vertical speed).This, owing to Parrot®, does not supply support over drivers design.Fuzzy control design is based on analyses over transitory response (Figure 5) and it is desirable that it includes the following features: • Total control this first stage of development is a combination of implemented control over platform (generally PID controllers) and a fuzzy controller designed to handle platform in transitory response.
• In this transitory stage, altitude and speed must be controlled because, given AR Drone weight and materials, it is possible to loss accuracy in position due to combined movements (transverse and longitudinal).
• Speed control includes monitoring of pressure sensor (i.e.ultrasound sensor at low altitudes) and control over accelerometer and gyroscope.This trying to obtain smoother movements in transitory stages.
• Vertical speed control is result of altitude control; therefore, this phenomenon allows platform navigation considering topography and other details of studied area.
Figure 6 presents in-block control diagram for previous features, where "h" is the data of altitude sensor and "v" is a data structure including speed in every axis.
Figure 6.In-block control diagram / Figura 6. Diagrama de bloques de control útil y potencia para ser llevadas a cabo.Para alcanzar la meta propuesta, el mecanismo de control de altura está basado en lógica difusa, ya que el control es altamente flexible y ajustable a condiciones cambiantes.
• El control de la altura es el efecto directo del control de la velocidad vertical; esta idea pretende la navegación de la plataforma con respecto a las depresiones del relieve.
La Figura 6 describe el diagrama de bloques de control para las características plan-

C. Simulation of global route planning algorithms
Currently, we are working on route planner simulation using navigation packages of the "Hector_Quadrotor" project, since it is the ROS base package for controlling AR Drone in Gazebo simulator (Figure 7).In general, simulation results do not consider environmental conditions because simulated object such as the AR Drone in ROS, do not support this kind of simulation in ROS.However, we found supports to link Matlab® with Hec-tor_Quadrotor through plugins, capable of evaluating engine voltage and the creation of wind vectors.
One of the first steps towards planner is a simulation of algorithms to obtain the shortest route.The basis of these algorithms are graphs to find corresponding nodes to the shortest route; they assume known environment (map) divided into squared cells.For implementation, graph must propagate over the map whilst the quadcopter is moving.Subsequently this results in translation of points of best route to the navigation space in simulator.We briefly present a description of implemented algorithms.

Wavefront propagation
This is also known as Breadth First Search (BFS) (Skiena, 1998).This algorithm starts in the root of the node tree (goal) and it explores every first neighbor node before it moves towards the ones in the following level.During propagation, the algorithm assigns weights putting "0" for goal, "1" for first neighbors, and so forth increasing in unary steps until the graph is complete.The best route is established by performing backward moving from the start to the end.The simulated algorithm in Gazebo considers the Von Neumann neighbors (Figure 8a).

Dijkstra algorithm
This algorithm takes the initial point and graph propagation to find the shortest route between every connected node; unconnected nodes are considered as unreachable (infinite distance) (Dijkstra, 1959).It uses a "0" for initial node and "infinite" for not visited nodes.Whilst the route expansion continues, the algorithm creates a set of unvisited nodes.For the actual node, the algorithm considers every neighbor as unvisited and it calculates distances; then, this distance is compared with the assigned value to select the lower.When every neighbor node of the actual one are covered, this actual is marked as visited and it is not checked again.If the destination node has been marked as visited, or if lower distance between teadas, en la cual h es el dato del sensor de altura y v es una estructura de datos que incluye la velocidad en cada uno de los ejes x, y, z.
Algoritmo de Dijkstra Este algoritmo toma un punto inicial y la propagación de un grafo para encontrar la ruta más corta entre todos los no- unvisited nodes is infinite, the algorithm ends (Figure 8b).

A* algorithm
It is one of the most used algorithms to determine routes from origin to destination points.A* implements heuristic processes in graph search, finding best route with lower computational costs compared with the previous ones (Hart, Nilsson, & Raphael, 1968).Its implementation is based on graph propagation over studied area and adding value (i.e.distance) from start point to the rest of the nodes "n".This process defines a function f(n) showed in (1) to evaluate which nodes are part of the route by considering actual cost until start point (g(n)) and the actual cost until goal (h(n)).
Given the analyses of previous algorithms, we present a qualitative contrast based on requirements of designed architecture for remote sensing system.We present this comparison in Table 1.dos conectados; los nodos no conectados son considerados a una distancia infinita (Dijkstra, 1959).El comportamiento de Dijkstra coloca un cero para el nodo inicial e infinito para los demás nodos marcados como no visitados.A medida que avanza la expansión de la ruta se crea un conjunto de todos los nodos no visitados.Para el nodo actual se consideran todos los vecinos no visitados y se calculan las distancias, luego se compara nuevamente la distancia con el valor asignado, para seleccionar el más pequeño.Cuando se han cubierto todos los vecinos del nodo actual, se marca el nodo actual como visitado y nunca será revisado de nuevo.Si el nodo destino ha sido marcado como visitado o si la distancia más pequeña entre los nodos del conjunto de nodos no visitados es infinita, entonces el algoritmo ha terminado (Figura 8b).

IV. Conclusions
Our proposal presented in this document is the result of previous literature reviews, finding gaps in research projects related with remote sensing systems with multi-rotor robots for precision agriculture.We present one possible solution as the design of the presented architecture, which pursues assistance in acquisition of crop information with similar reach of traditional techniques (aerial and satellite) but gaining access.
Development of current proposal involves combination of several reactive and deliberative behaviors; this in order to correctly adapt AR Drone quadcopter to local topography and, subsequently, obtain complete crop information.In consequence, to carry out these tasks we propose a non-linear altitude control based on fuzzy logic, aerial photography in crops in specific moments and points, and execution of efficient coverage plan from base stations.
Finally, for establishment of best coverage route, it is important to consider that AR Drone has a restricted activity time, and system must cover areas as extensive as possible.Hence, selection of algorithm for planning route coverage is mainly based on route with less turns, less predefined points, less visits, and a complete coverage.

Figure 3 .
Figure 3. Speed in "x", "y", and "z" axes for (a) a simulated scenario and (b) a real and controlled environment / Figura 3. (a) Velocidad en x,y, z para un entorno simulado y (b) para un entorno real controlado

Figure 7 .
Figure 7. Simulation and real model of AR Drone quadcopter / Figura 7. Modelo de simulación y modelo real de AR Drone