The drone market is still expanding. In the process of development, obstacle avoidance technology has become an important concern of R&D. The barrier technology is used as a guarantee to increase the safe flight of drones. Currently, the most popular obstacle avoidance technology has infrared sensors. Ultrasonic sensors, laser sensors, and vision sensors.
In recent years, the drone market has grown rapidly, and the obstacle avoidance technology as a guarantee to increase the safe flight of drones has been changing with the development of technology. During the flight, the drone collects information about the surrounding environment through its sensors, measures the distance and makes corresponding action instructions, thereby achieving the function of “obstacle avoidanceâ€.
At present, the most common obstacle avoidance technology for UAVs is infrared sensors, ultrasonic sensors, laser sensors, and vision sensors. Then why did Dali’s forward-looking obstacles first choose binocular vision? This is to start with the principles of each technology.
Infrared obstacle avoidance:We are no stranger to the application of infrared: from the remote control of TV, air conditioner, to the automatic door of the hotel, the infrared sensing principle is utilized. Specific to the application of the UAV to avoid obstacles, the common implementation of infrared obstacle avoidance is the "triangulation principle."
The infrared sensor includes an infrared emitter and a CCD detector. The infrared emitter emits infrared rays, and the infrared rays will reflect on the object. After the reflected light is received by the CCD detector, the reflection angle will be different due to the different distance D of the object. Different reflection angles will produce different offset values ​​L. Knowing that these data are calculated, the distance of the object can be obtained, as shown in the figure below.
Ultrasonic obstacle avoidance:Ultrasound is actually a kind of sound wave, because the frequency is higher than 20kHz, so the human ear can't hear it, and the directivity is stronger.
The principle of ultrasonic ranging is simpler than infrared, because sound waves are reflected by obstacles, and the speed of sound waves is known, so it is only necessary to know the time difference between transmission and reception, and the measurement distance can be easily calculated, combined with transmitter and receiver. The distance of the device can calculate the actual distance of the obstacle, as shown in the figure below.
Ultrasonic ranging is cheaper than infrared ranging, and the corresponding sensing speed and accuracy are also inferior. Similarly, since it is necessary to actively emit sound waves, the accuracy is also lowered with the attenuation of the sound waves for obstacles that are too far away. In addition, for objects that absorb sound waves such as sponges or in the case of strong winds, the ultrasonic waves will not work.
Laser obstacle avoidance:Laser obstacle avoidance is similar to infrared light, and it is also emitted by laser. However, laser sensors are measured in many ways, with triangulation similar to infrared, and time difference + speed similar to ultrasonic.
In either case, the accuracy, feedback speed, anti-interference ability and effective range of laser obstacle avoidance are significantly better than infrared and ultrasonic.
But here, note that whether it is ultrasound or infrared, or laser ranging here, it is only a one-dimensional sensor, can only give a distance value, and can not complete the perception of the real three-dimensional world. Of course, since the laser beam is extremely narrow, multiple lasers can be used simultaneously to form an array radar. In recent years, this technology has matured and is mostly used in self-driving vehicles. However, due to its large size and high price, it is not suitable for drones. .
Visual obstacle avoidance:The problem of how to "see" the robot is the computer vision that everyone often hears. The foundation is how to get 3D information from 2D images to understand the 3D world we are in.
A visual recognition system can generally include one or two cameras. A single photo only has two-dimensional information, just like a 2D movie. There is no direct sense of space. It is only by our own life experience that relies on "object blocking, near big and small". Therefore, the information obtained by a single camera and its limited, can not directly get the effect we want (of course, through some other means, assisted acquisition, but this is not mature, and there is no large-scale verification). Analogous to machine vision, the picture information of a single camera cannot obtain the distance relationship between each object and the lens in the scene, that is, the third dimension is missing.
Binocular stereo vision is like a 3D movie (the scenes seen by the left and right eyes are slightly different), which can directly bring a strong sense of space. Analog machine vision, from a single camera to two cameras, Stereo Vision can directly provide the third dimension of information, depth, which makes it easier to get 3D information. The most common example of binocular vision is our eyes: the reason why we can accurately pick up the cup in front of us and judge the distance of the car is because of the binocular stereo vision of both eyes, and the invention of 3D movies and VR glasses is also It is the application of binocular vision.
The basic principle of binocular vision is to use two parallel cameras to shoot, and then use a series of complicated algorithms to calculate the distance of a specific point according to the difference between the two images (parallax), and to generate depth when the data is sufficient. Figure.
Why is binocular vision stand out in drone applications?
In fact, each obstacle avoidance technology has its use in drones, but the application scenarios are different. Especially for forward avoidance obstacles, some technologies are not applicable.
Infrared and ultrasonic technology, because they need to actively emit light and sound waves, so there are requirements for reflected objects, such as: infrared rays will be absorbed by black objects, will penetrate transparent objects, and will be interfered by other infrared rays; and ultrasonic waves will be sponged, etc. Objects are also easily absorbed by the blade airflow.
Moreover, active ranging also creates problems with two machines interfering with each other. In contrast, although binocular vision also requires light, the requirements for reflectors are much lower, and the two machines do not interfere with each other at the same time, and the universality is stronger.
The most important thing is that the common infrared and ultrasonic waves are currently single-point ranging, and only the distance data in a specific direction can be obtained, and the binocular vision can obtain a relatively high scene in front of the small volume and low power consumption. The depth map of the resolution makes the obstacle avoidance function more development space, such as intelligent flight after obstacle avoidance and path planning.
Although laser technology can also achieve the function of binocular vision, it is limited by the development of technology. The current laser components are generally expensive, bulky, and high in power consumption. It is neither economical nor practical to be applied to consumer-grade drones.
Therefore, under the comparison of all parties, the binocular vision with high cost performance, simple principle, broad prospects and universal application stands out.
Five-way obstacle avoidance on Elf 4Pro
The five-way obstacle avoidance on the Elf 4Pro can be said to be a concentrated expression of the obstacle avoidance technology of the UAV. Therefore, we use the Elf 4Pro as an example. The obstacle avoidance before and after the P4P uses a binocular vision system. Compared to the Wizard 4, the data volume of the binocular vision only doubles. The combination of binocular vision + ultrasound is used to avoid obstacles to improve the stability and safety of flight in different environments. In the case of obstacle avoidance around the aircraft, infrared obstacle avoidance is adopted.
Visual odometerIt is worth mentioning that the binoculars of the lower view use the technology of visual odometer (VO).
In a simple way, the visual odometer is "the three-dimensional position of the object in the field of view is reversed by the left and right images," so the speed and height can be simply measured compared to the optical flow + ultrasonic technology. The visual odometer can also be constructed. The three-dimensional model of the ground, and through the continuous image, track the relative movement of itself and the environment, and estimate its own motion. Accurately measure the relative position of itself and the ground.
Although the data processing capacity of the visual odometer is several times that of the optical flow method, it is because of its introduction that the Elf 4 and the Elf 4Pro can clearly control their position in the GPS-free room, thus achieving stable stability. Hovering and not appearing high.
With the visual odometer, combined with the map reconstruction of the stereo vision before and after, the drone grasps the position of the obstacle and the position of the machine at the same time. At this time, it is easy to drive the motor to bypass, a complete avoidance. This is how the barrier function is implemented.
Flight Autonomy SystemThe five-way obstacle avoidance of Elf 4Pro is not independent. The binocular vision, the lower-view binocular + ultrasound and the left and right infrared obstacle avoidance of the front and rear view together constitute the FlightAutonomy system of Dajiang. This system gives the Elf 4Pro the ability to remember the three-dimensional environment. The front and rear binoculars and the lower binoculars can construct and record the surrounding terrain in real time, specifically, local mapping and global mapping.
The local map allows the drone to build and remember the three-dimensional environment within a few tens of meters around it, thus achieving the functions of "pointing flight", because only knowing the direction of flight and the position of the fuselage is definitely not enough, only the ability to remember the surrounding terrain changes. In order to complete the task of "planning the route in the specified direction and bypassing".
The global map is a record of the terrain that has passed through the entire flight. Although the accuracy is not as good as the local map, it can help the drone realize the functions of “smart returnâ€: when the drone accidentally flies to the back of the building, it leads to remote control. When the signal is lost, the smart return function allows the drone to return to the original route within one minute. If the remote control signal cannot be connected within one minute, it will return straight. This memory of the environment around the flight path is another embodiment of the Smart 4Pro intelligence.
Difficulties in implementationFrom the conception to the realization of the obstacle avoidance function, almost every step of the process is followed by numerous problems. Just writing an effective visual recognition or map reconstruction algorithm is only the first step. It is really difficult to run smoothly on a platform such as a drone that has limited computing power and power consumption. The place. Especially on the Elf 4Pro, not only the amount of binocular vision data is doubled directly compared to the Elves 4, but also the endurance is not affected, which is very difficult.
In addition, how to deal with the boundary of the function is also a problem, for example, binocular vision can work under the condition of good sight, then when there is dust blocking? This requires constant experimentation and trial and error, and continuous optimization algorithms to ensure that all functions can work normally under various scenarios, without giving wrong instructions.
As the general trend of the UAV products in recent years, the most direct benefit of the "obstacle avoidance function" is that some of the collisions caused by human negligence can now be avoided through the obstacle avoidance function, which ensures the drone flight. At the same time of safety, it also avoids damage to the property of the surrounding people, and the threshold of the flying drone is further reduced.
In the long run, drones want to spread to agriculture, construction, transportation, media and other fields. "Intelligence" is definitely the only way. After all, only when the flight function is intelligently controlled, there is a surplus to satisfy. The needs of different industries. Nowadays, a series of "smart flight" functions derived from the "obstacle avoidance function" is undoubtedly one of the staged manifestations of "UAV intelligence".
Switching Residual Current Transformer
Switching Residual Current Transformer,Switching Transformer,Open Residual Current Transformer,Busbar Open Residual Current Transformer
Zibo Tongyue Electronics Co., Ltd , https://www.tongyueelectron.com