Vision-Based Autonomous Landing and Charging System for a Hexacopter Drone

At present, several approaches for automating the landing process of a Drone on a landing platform have been studied. A common method is to use vision-based autonomous landing technique for controlling the behaviour of the Drone [8-10]. Patruno et al. [8] present a vision-based system for the pose estimation of the Drone. Their methodology was based on a coarse-to-fine approach to search the target markers. A vision-based system for the position and attitude estimation of the Drone is presented in the study [9], a “Hough transform” used to extract the straight lines in the target images. Those lines are then used with the captured images to estimate the position and attitude of the Drone. The main research groups involved in the development of vision-based autonomous landing systems are reviewed and presented in the study [10], where the details of each algorithm and system is discussed.

In order to recognize the landing platform and assist Drones to land safely, artificial markers (helipad) are widely used. Liu et al. [11] designed a new type of landing pad, which includes several ArUco markers surrounded by a circle. Each marker is composed of a wide black border and an inner binary matrix that defines its identifier. An on-board vision system for autonomous take-off, hovering and landing of MAV is presented in the study [12], the system is based on a single image of a typical landing pad which consists of the letter “H” surrounded by a circle. Design and implementation of a real-time, vision-based landing algorithm for an autonomous helicopter is reported in the study [13], the system is also based on a helipad that consists of the letter “H”. The same mark is used in the study [14]. A visual framework is developed to allow the autonomous landing of a drone on a platform using a single camera. Two methods for automatic detection of landing pads were proposed in the study [15], the landing pads. The landing pads are marked by traditional characters “H” and “X” inside of circles. A custom ArUco marker was proposed in the study [16] which consists of a small marker nested inside a bigger one.

A popular way to achieve a high-precision landing is to use infrared technologies. Gui et al. [17], used four infrared light-emitting diode lamps to positively identify the landing zone. More application of autonomous landing algorithm, that allows the vehicle to land robustly and precisely onto a heaving platform is presented in the study [18], where IR MarkOne beacon and IR Pixy camera were utilized. Hayajneh and Badawi [19], employ the IRLock sensor to track a MarkOne beacon, together with a rangefinder to detect the vertical distance to the landing position. The IRLock sensor is based on the Pixy vision sensor, while the MarkOne beacon is simply an array of infrared LED’s. A precision landing system that consists of an infrared camera and an IR light beam is proposed in the study [20], and a more complex system with an array of 2×72 LEDs, Raspberry Pi, and NoIR camera is presented in the study [21].

Currently, the most common approach is to combine different technologies such as visual markers with infrared technologies. An array of 264 infrared LED and ArUco marker, proposed in the study [22] were used in designing the landing pad. Such design makes the marker visible for a conventional camera even at poor light conditions. Wang and Bai [23] used a square plate as a special landing pad. On the top side of the square plate, an ‘H’ sign is attached at the center and surrounded by 4 red LEDs. These LEDs are used to guide the Drone from long distances. A landing pad consisting of a fractal ArUco marker and a MarkOne beacon is presented in the study [24]. In this research twenty infrared light-emitting diodes and eight barcodes are used to feature a special landing and Conductive Copper Sheet.

In order to increase the endurance and overall flight time of Drones in a real time, several approaches have been proposed to autonomously recharge their batteries during long duration missions. A hot swapping technique is presented in the study [25], such technique is partially autonomous, where Drone is required to fly back to the ground station to replace its battery either manually or automatically. Another technique which is completely autonomous was proposed in the study [26]. Therein, the author presents the automated battery swapping system which includes the design concept of the battery swapping mechanisms and a precise landing technique. Dale [27], describes the design, testing, and construction of an autonomous recharging station for quadrotor using contact-based charging stations. Such method requires a precise landing system and a mechanical system that is able to strongly bond the electrodes for the conductivity.

Autonomous landing and charging systems using onboard vision-based system requires robust algorithms. In order to deal with the control problem of such systems; A number of control algorithms were investigated such as proportional-integral-derivative (PID) [28], model predictive controller [29], and backstepping controller [30]. More advanced controllers were also used such as neural network controller [31], fuzzy logic controllers [32], multi-level fuzzy logic controller [33], and deep reinforcement learning algorithm [34]. Although many control techniques were presented, still designing a high-performance controller for landing and charging Hexacopters is a challenging issue.

In numerous researches focused on achieving precise drone landings, the level of accuracy typically depends on the performance of the camera and the presence of a visible ground marker, vision-based method and development has been widely investigated [35, 36]. Despite the high precision in landing, adaptability, and many advantages of the vision-based autonomous system, still, there are some essential challenges associated with the vision system. such as sensitivity to environmental changes [37], computational complexity, and the quality of cameras system [38]. At the other hand, artificial marker (pads), have also their benefit and challenges, compared to the vision system, Marker-based systems are less sensitive to environmental changes like weather conditions or lighting variations, providing a more robust solution in diverse operational settings, [39]. For adaptability, artificial markers offer less adaptability to dynamic environments or changing conditions compared to vision-based systems, as they depend on fixed markers at predetermined locations, [40]. A combination of both approaches, where feasible, may provide a comprehensive solution that leverages the strengths of each method which is adopted in the research. Moreover, using a combination of visual markers and infrared markers can offer versatility. This approach allows the drone to adapt to changing lighting conditions, and increase the success rate of landing process during different weather conditions. In general, the choice between such systems depends on the specific requirements of the application, the operational environment, and the capabilities of the drone's sensors. Adverse weather conditions, such as strong winds, rain, or fog, can significantly impact drone operations. Winds and rain issues are considered as one of the main operator tasks for setting up a pre-flight planning. The current research addresses the fog issue by combining two techniques while searching the landing area; it mainly focusses on the use of infrared technology. Flying Drone at high altitude presents some unique challenges; such as payload and flight duration. Altitude limits for the current research is limited to 10m, which required a light weight vision system and requires less time to land Drone. On the other hand, any extra wight related to an advanced computer or cameras requires more power which reduces the flight duration.

In this paper, a hierarchical vision-based autonomous landing algorithm (HVALA) based on Otsu thresholding method and LOG operator is proposed. This algorithm combines two approaches to guarantee a successful landing and improving the charging process.

4.1 Camera matrix model

To describe the correspondence between a point in the world coordinate system and a point in the two-dimensional image plane, let $\left(X_w, Y_w, Z_w\right)^T$ be the space coordinates of a point $P_w$ in world coordinate system visible to the camera. Meanwhile, let $\left(X_c, Y_c, Z_c\right)^T$ be the image coordinate of the same point $P_C$ in 3D camera coordinate system and $(\tilde{\mathrm{x}}, \tilde{\mathrm{y}})^T$ be the image coordinate of the same point $\tilde{P}_c$ in the twodimensional image plane. Working in homogeneous coordinates, the relationship between a point $P_c$ in the threedimensional camera coordinate system and a point $\tilde{P}_c$ in the two-dimensional image plane using the intrinsic matrix $k$ of a camera can be described in a matrix form as follows:

$\tilde{P}_{\mathrm{c}}=P_o P_c$                   (1)

where, $P_o$ is a camera matrix defined as $P_o=k[I \mid 0]$, and the intrinsic matrix $k$ is

$k=\left[\begin{array}{ccc}f_x & 0 & c_x \\ 0 & f_y & c_y \\ 0 & 0 & 1\end{array}\right]$                  (2)

Then

$\left[\begin{array}{l}\tilde{\mathrm{x}} \\ \tilde{\mathrm{y}} \\ \tilde{\mathrm{z}}\end{array}\right]=\left[\begin{array}{cccc}f_x & 0 & c_x & 0 \\ 0 & f_y & c_y & 0 \\ 0 & 0 & 1 & 0\end{array}\right]\left[\begin{array}{c}X_c \\ Y_c \\ Z_c \\ 1\end{array}\right]$                      (3)

Since the point $P_w$ is located in the world coordinate system, then, an additional transformation is needed to relate the point $P_w$ from the world reference system to the camera reference system. The transformation is accomplished by a rotation matrix $R_{3 \times 3}$ and translation vector $t_{3 \times 1}$ (extrinsic parameters). Therefore, the coordinates of the point $P_c$ given the coordinates of the point $P_w$ can be computed as follows:

$P_c=\left[\begin{array}{cc}R & t \\ 0_{1 \times 3} & 1\end{array}\right] P_w$                            (4)

Substituting Eq. (4) in Eq. (1) and then in Eq. (3) will give us the following formulas:

$\begin{aligned} & \tilde{P}_{\mathrm{c}}=P_o P_c \\ & \tilde{P}_{\mathrm{c}}=k[I \mid 0]\left[\begin{array}{cc}R & t \\ 0_{1 \times 3} & 1\end{array}\right] P_w\end{aligned}$                      (5)

$\left[\begin{array}{c}\tilde{\mathrm{x}} \\ \tilde{\mathrm{y}} \\ \tilde{\mathrm{z}}\end{array}\right]=\left[\begin{array}{llll}c_{11} & c_{12} & c_{13} & c_{14} \\ c_{21} & c_{22} & c_{23} & c_{24} \\ c_{31} & c_{32} & c_{33} & c_{34}\end{array}\right]\left[\begin{array}{c}X_w \\ Y_w \\ Z_w \\ 1\end{array}\right]$                  (6)

The $c$ matrix in Eq. (6) is called complete camera calibration matrix, and the formula is used to map any point $P_w$ in world reference system to the image plane. Accordingly, the exact pixel coordinates of the detected object returned by the camera can be obtained by dividing the first two coordinates of $(\tilde{\mathrm{x}}, \tilde{\mathrm{y}}, \tilde{\mathrm{z}})^T$ by the third coordinate $(u=\tilde{\mathrm{x}} / \tilde{\mathrm{z}}$, $v=\tilde{\mathrm{y}} / \tilde{\mathrm{z}})$ [7].

4.2 Camera calibration

The open-source library OpenCV and classical black-white chessboard are used to estimate the extrinsic and intrinsic camera parameters. In our experimental setup, we calibrated two different cameras: IMX322 and Pixy cameras. We decide to use an 8×6 chessboard as a real-world object for the calibration process. Chessboards are commonly used for camera calibration because they provide a regular pattern of high-contrast features, which facilitates the calibration process. In order to achieve the camera calibration with a balance between having enough calibration points and avoiding unnecessary complexity, we have used am 8×6 chessboard which provides enough points to accurately estimate the intrinsic parameters of the camera. Using a larger chessboard is possible, but that would at the expense of computational complexity, moreover such size choice is very common in the field of computer vision for camera calibration and hence ensure the compatibility across different software tools as well as vision libraries (such as OpenCV, which provides functions specifically designed for detecting corners in an 8×6 chessboard, and it has been used in our experimental studies), although In some cases, especially when dealing with highly specialized applications or specific camera properties, other sizes of chessboards might be chosen for optimal results.

Figure 2 shows a chessboard image with the corners detected for both cameras. We used the OpenCV built-in function cv2.calibrateCamera() to calculate the camera matrix and distortion coefficients. Tables 1, and 2, show our result for both cameras. The camera matrix includes the intrinsic and extrinsic parameters of the camera. The intrinsic parameters of the camera composed of the focal lengths $\left(f_x, f_y\right)$ and the optical center of the camera $\left(c_x, c_y\right)$. Extrinsic parameters $(t, r)$ are related to the translation and rotation vectors, which are used to convert images from camera coordinate system to world coordinate system.

2.png

Figure 2. Chessboard image (a) chessboard image with corners detected by IMX322 camera, (b) chessboard image with corners detected by Pixy camera

Table 1. Distortion parameters of the IMX 322, and Pixy camera

Table 2. Intrinsic parameters of the IMX322, and Pixy camera

There are two major kinds of distortion caused by camera Radial distortion causes the straight lines in a real world to be curved when they appeared in the image, while tangential distortion causes the distance between the object and the viewing point to be closer than expected. Distortion coefficients include the radial distortion $\left(k_1, k_2, k_3\right)$ and tangential distortion $\left(p_1, p_2\right)$, in other words, the above five parameters known as distortion coefficient must be found together with above mentioned parameters. Such calibration is essential in our proposed system since the autonomous landing and charging process depends on the correct detection of the landing platform (Barcode, IR, and Conductive Copper Sheet) where the onboard camera is the responsible for detecting the regions of the IR-LEDs. The correct identification of camera parameters, ensure the correct calculation and detection of the landing platform and the center of the four regions at the edge (IR1, IR2, IR3, IR4). Data from the calibration process are obtained and used for the IR-LEDs detection and position estimation presented in subsection 4.3 and 4.4.

4.3 Pose estimation of the drone

As shown in Figure 3, a landing pad with five groups of infrared LEDs are used as a source for Drone pose estimation. In our experimental setup, Otsu's thresholding method is used to determine the infrared LEDs regions in the image. We then apply the Laplacian of Gaussian (LOG) filter to detect the bright blobs in the image. Since there are five bright bobs in the image, we use cv2.findContours() OpenCV built-in function to find the number of contours in the Image and then we calculate the moments for each contour using cv2.moments(). The centroid of each contour is then calculated by the formula $C x=M 10 / M 00$ and $C y=M 01 /$ $M 00$, where $C x$ is the $\mathrm{x}$ coordinate and $C y$ is the y coordinate of the centroid. Meanwhile, M10 is the 1st order spatial moment around x-axis, $M 01$ is the 1 st order spatial moment around y-axis, and $M 00$ is the 0 th order central moment. The centroid of the four regions at the edge (IR1, IR2, IR3, IR4) are then calculated $\left(u_a, v_a\right)$ and compared with the center of IR5 using $\left(\left(u=a b s\left(u_5-u_a\right),\left(v=a b s\left(v_5-v_a\right)\right)\right.\right.$, if the difference is less than a certain threshold, then the four edges will be considered as a searching area and their center will be tracked. The cv2.solvePnP() OpenCV built-in function is used to return the translation and the rotation vectors. The translation and the rotation vectors are then converted to the rotation matrix by the cv2.Rodrigues() OpenCV built-in function, which contains the roll, pitch, and yaw angle of the IR5 marker relative to the camera in the marker's coordinate system. Later, we calculated camera's position and attitude relative to the IR5 marker.

3.png

Figure 3. Landing pad with five groups of IR-LEDs

The coordinate system of the camera is aligned with the drone’s coordinate system so that the position and attitude of the camera frame in respect to the IR5 marker is also the position and attitude of the Drone relative to the same marker. The position and attitude of the Drone are then sent to the Pixhawk flight controller so that it can hover directly above the center of the IR5 marker before landing. The landing process is then initiated by activating the Pixy cameras. Pixy cameras are trained to detect the barcodes and then land Drone at the correct charging plate.

4.4 Vision algorithm and control strategy

In order to achieve a high-precision landing task, and then initiate the charging process, a hierarchical vision-based autonomous landing algorithm is proposed in this research. The flowchart of our algorithm is presented in Figure 4.

Initially, Drone is required to take-off and fly toward the first waypoint using the GPS module, and then autonomously locate and recognize the candidate regions of the IR-LEDs in the image. If the Raspberry Pi does not detect any bright blobs in the image, it commands the Drone to fly towards the next waypoint and then polls the camera again until it detects five blobs in the image. If this is the case, Raspberry Pi will command the Pixhawk flight controller to fly Drone towards the WP5 and descend from there. During this period, Raspberry Pi will extract the center of each region. The center of the four regions at the edge (IR1, IR2, IR3, IR4) are then calculated and compared with the center of IR5. If the difference between them is larger than nine pixels (such threshold is driven by several experimental studies designed and measured for this purpose. The measures consider the dimension and placement of the conductive sheet on the Drone and ground station. The threshold is determined in order to cover the optimal match of the charging plates as well as the maximum possible placement of the conductive sheet subject to successful charging based on the actual dimension of the designed ground station), Raspberry Pi will command the Pixhawk flight controller to change flight mode to position hold and then Yaw Drone 15 degree toward north.

4.png

Figure 4. A flow-chart of the hierarchical vision-based autonomous landing algorithm (HVALA)

The previous process will be executed continuously until the result is below nine pixels. If this is the case, pose parameters of the main camera with respect to the IR5 marker are calculated, and then results will be sent to the Pixhawk flight controller to position Drone above the center of the IR5 marker before landing. During this period, altitude of Drone is calculated. If the result is less than two meters, Pixy cameras are activated. Two Pixy cameras are trained to detect the barcodes and then land Drone at the correct charging plate. The first camera is attached to the Arduino Nano board to look for the barcode values (4, 12, 2, 14, 0), while the second camera is attached to another Arduino Nano board to look for the barcode values (4, 12, 6, 10, 8). If this is not the case, Raspberry Pi will command the Pixhawk flight controller to Yaw Drone 45 degrees and then polls the camera again to check if the barcodes (4, 12) are detected by both cameras. Once this occurs, Raspberry Pi will land Drone and initiate the charging process.

4.5 System pre-requisite considerations and constrains

The proposed system introduces a novel autonomous landing and charging paradigm with combined technologies. Still, as in any other developed system, some constrains and operational parameters contribute to its functionality and effectiveness; such keys are summarized as follows:

4.5.1 Takeoff constrains

Within the used size of landing platform which is relatively small, the drone take off is limited to an altitude of less than 10 meter in order achieve a successful detection of the landing platform. Such constrains can be adjusted upon the variation of the landing platform dimension which is considered for further investigation in future work.

4.5.2 Infrared visibility and barcode detection

Leveraging infrared technology, the visibility of barcodes is constrained to a maximum distance of 2 meters. Beyond this range, the Pixy cameras cease to effectively observe the barcodes. Upon reaching a height of 2 meters above the ground, the Pixy cameras come into play, detecting barcodes.

4.5.3 Normal flight control parameter

The drone adheres to standard flight control parameters, taking into account factors such as wind speed, temperature, and other environmental conditions to ensure stable and controlled flight.

4.5.4 Operation constrains

Operational considerations include a limited flight duration (i.e within the current capability of Drone technology), emphasizing the importance of optimizing energy consumption and flight efficiency. To activate the charging process, a correct alignment of the polarity of conductive copper sheet is required.

As shown in Figures 7, 8, and 9, several experiments have been performed to measure the actual accuracy of the autonomous landing and charging algorithms. In the first set of experiments, the Drone is placed at 8 meters away from the landing platform and then commanded by the Raspberry Pi to take-off to 10m altitude. When Drone reaches the 10m take-off altitude, it flies toward the landing platform based on the GPS module. It autonomously follows a predefined trajectory which is defined via five waypoints (WP1, WP2, WP3, WP4, WP5). Waypoint five indicates the position of the landing platform, while the others represent a horizontal square path surrounding the position of the landing platform that Drone might initially follow. The distance between the waypoints (WP1, WP2, WP3, WP4) is set to 6 meters, while the WP5 is located at the center of the square.

As soon as Drone reaches the first waypoint, it starts the searching process by enabling the onboard vision system. Once the main onboard camera recognizes the landing platform and detects the infrared LEDs, Drone is commanded to move towards WP5 and starts to descend from there slowly. On other hand, if Drone does not detect the infrared LEDs, it will fly to the next waypoint and starts another landing attempt from scratch (i.e flying from any waypoint to WP5 and then start landing). If this is the case after visiting WP4, Raspberry Pi will command the Pixhawk controller to fly the Drone back to the initial point from where it starts another searching attempt from 8m altitude.

In the second set of experiments and during the landing process, a threshold of 2m is set above the position of the landing platform to decide or deny the last phase of the descent process. Pixy system (1) (i.e pixy camera and Arduino) is programmed to detect the barcodes (4, 12, 2, 14, 0), while Pixy system (2) is programmed to detect the barcodes (4, 12, 6, 10, 8) and then land the Drone at the conductive copper sheets. At this step, Raspberry Pi will command the Pixhawk flight controller to change the flight mode to position hold. Position hold is one of the flight modes that pause the descent process and automatically attempts to maintain the current location, heading and altitude of the Drone. During this period, if the polarity of the conductive sheets on Drone does not match the polarity of the charging station, Raspberry Pi will command the Pixhawk controller to Yaw Drone 45 degree until they get matched. Once this process finished, Raspberry Pi will land the Drone and initiate the charging process.

In case that the conductive copper sheet of the ground station are lost, Raspberry Pi will command the Pixhawk controller to fly the Drone back to WP1 from where it starts another landing attempt from scratch. In addition to the aforementioned precautions, specific operational protocols have been established to enhance safety and reliability during the hexacopter's various phases of operation, discussed briefly in Table 6.

Table 6. Main safety measured implemented during the hexacopter's various phases of operation

These additional measures, spanning take-off, flight initiation, landing configurations, and low battery scenarios, contribute to a comprehensive safety framework, reinforcing the reliability and autonomy of the hexacopter's operational protocols.

Figure 7 (a) shows an aerial view of the landing platform used in our experiments. The image is taken in flight from the main downward-pointing camera without the IR filter. Figure 7 (b) shows the actual path that is followed by the Drone using the GPS and the main downward-pointing camera with the IR filter. As expected, Drone moved towards WP5 as soon as it reached WP3. This indicates that the landing platform gets in view before reaching WP4. A photo is taken during this period and is shown in Figure 8 (a). From this point, Drone starts to descend to a lower height slowly, Figure 8 (b). After each time that Drone descends, the system checks if the height of the Drone relative to the landing platform is above 2m or not.  Once the Drone’s height is below 2m, Figure 8 (c), the Pixy cameras are activated. From this point and onwards, Drone slowly lands automatically on the charging platform, a successful autonomous landing of the Drone is shown in Figure 8 (d). Meanwhile, the position and the attitude of the main onboard camera with respect to the LED’s breadboard is shown in Figure 9. Figure 9 (a) shows an aerial view of the infrared LEDs at an altitude of 8m, while Figures 9 (b), (c), and (d) show three trials while Drone descends on the charging platform. In each experiment, we recorded the landing position of the Drone and measure the distance between the center of the landing pad (IR5) and the center of the main onboard camera. A small error of 4.4cm on average was observed in the landing process. The deviation of up to 4.4cm is deemed acceptable in this application. The drone is programmed to initiate the landing sequence when the error between the center points of IR1, IR2, IR3, and IR4 and the center of IR5 falls below 4.5cm (equivalent to 9 pixels). The LEDs have been strategically placed on a (9×9)cm breadboard, creating a circular arrangement with a radius of 6.36cm. In the charging initiation process, if the drone comes within a distance less than 6.36cm from the central point, it is recognized as a successful condition for triggering the charging process. This error might be solved by increasing the width of the conductive copper sheet, since the width of the conductive copper sheets attached to the drone is 1cm while the width of conductive copper sheet at the landing platform is 5cm. As a result, any error less than 5cm is acceptable for this application. On the other hand, any error larger than 5cm and less than 6.36cm requires to increase the width of the copper sheets attached to the landing platform from 5cm to 7cm, which leads to a higher landing success rate. In addition, the last phase introduces an additional time overhead in the landing process. Trial results are listed in Table 7.

7.png

Figure 7. Planned mission, (a) Aerial view of the landing platform without the IR filter, (b) Actual path that is followed by Drone using the GPS and the main downward-pointing camera with the IR filter in one of the tests

8.png

Figure 8. Landing flight tests, (a) Drone is hovering over the landing platform at 10m altitude, (b) Drone is hovering over the landing platform at 5m altitude, (c) Drone is hovering over the landing platform at 2m altitude, (d) Drone after landing on the ground platform

9.png

Figure 9. LEDs detection and position estimation process, (a) Aerial view of the LEDs at an altitude of 8m, (b) The first trial that indicates the position and the attitude of the camera with respect to the infrared LED’s breadboard, (c) The second trial that indicates the position and the attitude of the camera with respect to the infrared LED’s breadboard, (d) The third trial that indicates the position and the attitude of the camera with respect to the infrared LED’s breadboard

Table 7. The results obtained during the final phase of the landing process using the Pixy cameras

In this research, we have presented the design and implementation of a real-time vision-based autonomous landing and charging system for a Hexacopter Drone. The system aimed to autonomously recharge drained battery and extend the overall flight duration. It based on the infrared LEDs detection and marker recognition. A custom-built hexacopter is used to demonstrate the proposed system and being equipped with a telemetry radio, a GPS receiver, a Pixhawk flight controller, electronic speed controllers, BLDC motors, propellers, barometer, magnetometer, accelerometer, gyroscope, three cameras, Raspberry Pi, balancing charging board, and LiPo battery. A novel landing platform with twenty infrared light-emitting diodes and eight barcodes was carefully designed and used in this research. A hierarchical vision-based autonomous landing algorithm (HVALA) based on Otsu thresholding method and Laplacian of Gaussian (LOG) operator was implemented in this research. The presented algorithm composed of two phases. Both are working sequentially to manage the landing and charging process. During the first phase, the LEDs were observed by the main camera. Algorithm was designed to keep the center of the infrared LED’s breadboard in the center of the image. In the second phase, Once the Drone is 2m above the ground, barcodes were observed by two Pixy cameras. The main strategy of the algorithm in this stage is to align the polarity of the main onboard battery with the charging system and then trigger the charging process. Thereby Drone can continue the mission after being recharged without any external interruption. We demonstrated the feasibility of the proposed algorithm through a series of autonomous flights experiments. The results obtained from the real-time flight experiments have shown that an autonomous landing to within 4.4cm in radius was achieved while introducing an additional time overhead in the landing process.