The core technology of unmanned factory is three-dimensional image modeling and spatial positioning

The previous article refers to the Han O Sinox unmanned factory, certainly more advanced than the current automation production line. We see the Automation factory robotic arm is the automation component, but only a relatively fixed operating mode, the operating position is very fixed, only some simple mode. For example, the screw, the estimated offset, the manipulator is powerless, that is, the manipulator is not too smart, the first is no eyes, and then can not arbitrarily change position. This is about image recognition modeling and space positioning techniques. I think these techniques should be very avant-garde, why should we say it? Not afraid of being imitated by others? Not afraid, I am not afraid of others to master, I am afraid that some companies take these things to apply for patents, such as the recent Google actually applied to the robot implant character this patent, it is ridiculous, robot can not have character? Also to pay Google royalties? I have the avant-garde technical details of the public, such as the matrix computer, see you how to apply for patent. Unmanned factories are not patented, and technology has long been public, and here is one. A company can not rely on royalties to survive, such as the recent Blizzard Warcraft game is also declining.

Image three-dimensional modeling actually said is the image recognition technology, to the object appearance recognition, may need multiple angle plane plan to establish three-dimensional model, people say holographic photography may go further, but the current three-dimensional graphics, similar to the 3d film we look at, estimated to be enough to identify the object, such as screw, the appearance is hexagonal or four corners, Thickness is also the height of how many millimeters, through the three-dimensional model can be identified tens of thousands of different nut. Input image and calculate the generation of three-dimensional model, which is the result of image recognition, the computer processing image is still very slow, we should develop three-dimensional image modeling algorithm, and the module and hardware, the design of a dedicated hardware chip after tens of thousands of parallel computing, the instant completion of image recognition. This is a bit like CAD mapping, but it is not manual painting, but using multiple images to generate the model, just like the model printed out, it is reverse, input graphics, generate models. At present, the word recognition technology is more mature, but the pattern recognition is not mature enough, because not only the floor plan. It is of little significance to restore the line graph in floor plan. Only a number of angle images to restore the stereoscopic map is the key, of course, will not be applied to ordinary users. At present, 3d movies can let the eyes see 3d graphics as long as two angles of view pictures. In fact, image recognition needs to be a three-dimensional model of the object, not just the mesh to recognize what the object, for the robotic arm, to recognize the object is not enough, but also to know the shape and size, so as to precisely locate.

Now the robotic arm is preceded by a camera, ingesting object images, and modeling in three dimensions, knowing whether the object is a screw or nut, size and position.

The last is to locate, the arm can clamp the nut, the arm moved to the top of the nut, a bit like missile guidance, arm side to move side to see, calibration position, put the clip in the nut is just below, and then down, tightly clamped, of course, clamp more tightly can also be precisely controlled, and the current mechanical arm clamping force is fixed. In fact, positioning technology can also be used radar technology, radar is actually emitting radio waves, electromagnetic waves, and then receive reflection back, calculate the round-trip time, you can calculate the distance. The bat is the echo location is the throat issued ultrasonic waves encountered obstacles back to the eardrum can be sensed, ultrasound is UHF sound. Can precise positioning also be achieved using radar technology and echolocation technology? Clamp the nut and move it to the screw, tighten it and complete the screw process. This process can be done with eyes and hands, without the need for radar and echo positioning. Since the eyes are so good, why not use them? As long as there are two cameras installed in front of the robot arm, two angles are recognized simultaneously, can calculate the exact distance between the object and the manipulator, and the operation can be calculated by the side moving edge to achieve precise control. At present, the computation of image recognition is still too large for the computer, and it is very important to develop an image-modeled chip to be competent for this work, which is widely used in industry, agriculture and transportation. The distance between the object and the camera can be calculated by using the three-dimensional model generated by two angle image modeling and the distance of two cameras to complete the space positioning. It seems that animals have 2 eyes, for a reason.

The main technology of unmanned factory is image modeling and space positioning, it does not need speech recognition, and the gap with the robot. Foxconn says developing robots to replace workers, but factory robots do not have language input and output, at least factory robots do not need to send and receive voice commands. So even if the current speech recognition and output technology is not mature, the development of the robot obstacle is image recognition and spatial positioning. There is no high-speed image recognition chip, the development of industrial robot cost is very high, Foxconn has no chip development technology, can only develop simple robotic auxiliary assembly. GPU image processing chips may contribute to the development of image recognition chips.

And the CAD software can print out the model, but can not restore 2 images of the model. Therefore, there is no mature image recognition algorithm. 3D image recognition modeling should begin with a 2D floor plan.

The core technology of unmanned factory is three-dimensional image modeling and spatial positioning