Return on robotics and Servo Mechanism
This definition implies that a device can only be called a â € € œrobotâ if it contains a movable mechanism, influenced by the detection, and action planning and control components. This does not imply that a minimum number of these components must be implemented in software, or be modified by the â € € œconsumerâ using the device, for example, the behavior of motion can be been hard-wired into the device by the manufacturer.
Â
Therefore, the definition presented as well as the rest of the material in this part of the reserve, covers not only œpureâ € â € â € robotics € œintelligentâ only robots, but rather the limited broad domain of robotics and automation. This includes â € € œdumbâ robots, such as metal and woodworking machines, â € € œintelligentâ washing machines, dishwashers and pool cleaning robots, etc. All these examples have sensing, planning and control, but not often individually separate components. For example, detection and planning behavior of the pool cleaning robot have been integrated into the mechanical design of the device, the understanding of human development.
Â
The robot is in a very large extent, all about systems integration, achieving a mission a mechanical device powered through a â € € œintelligentâ integration of components, many of which are shared with other domains, as systems and control, computer character animation, machine design, computer vision, artificial intelligence, cognitive science, biomechanics, etc. Furthermore, the limits of robotics can not be clearly defined, as also the â € € œcoreâ ideas, concepts and algorithms are being applied in an increasing number of applications œexternalâ € â €, and, conversely, the core technology in other domains (vision, biology, cognitive science or biomechanics, for example) are becoming increasingly vital components in modern robotic systems.
Â
This part of the WEBook makes an effort to define exactly what is referred to the base material of the domain of robotics, and to describe in a coherent and motivated. However, this structure chosen is only one of many possible â € € œviewsâ you may want to have in the domain of robotics.
Â
In the same vein, the aforementioned â € € œdefinitionâ of robotics is not intended to be definitive or final, and is only used as a general framework for structuring the different chapters
Â
Components of robotic systems
Â
Â
Â
Â
Â
Â
Â
Â
This figure shows the components that are part of all robotic systems. The purpose of this section is to describe the semantics of terminology used to classify WEBook chapters: œsensingâ € â €, â € œplanningâ €, â € œmodelingâ €, â € œcontrolâ €, etc.
Â
The real robot is a mechanical device (â € œmechanismâ €) moving around in the environment, and in doing so, physically interacts with this environment. This interaction involves the exchange of physical energy, one way or another. Both the robot mechanism and the environment may be the â € € œcauseâ physical interaction through â € œActuationâ € or â € € experience œeffectâ of interaction, which can be measured through œSensingâ € â €.
Â
Robotics as an integrated control system of interaction with the physical world.
Â
Sensing and actuation are the physical ports through which the â € € œControllerâ the robot determines the interaction of mechanical body with the physical world. As mentioned before, the driver can, at one end, is the program, but on the other end all that it can also be implemented in hardware.
Â
Inside the Controller component, several sub-activities are often identified:
Â
Modeling. The relations of input-output All control components may (but need not) be derived from information that is stored in a model. This model can take many forms: analytical formulas empirical search tables, fuzzy rules, neural networks, etc.
Â
The name â € € œmodelâ often leads to heated discussions between different research œschoolsâ € â €, and WEBook not interested in taking a stand in this debate: the WEBook, â € € œmodelâ be understood with minimal semantics: â € œany information used to determine or influence relations input-output components of € Controller.â
Â
The other components discussed below can have all the models inside. A € â € œSystem Models can be used to tie multiple components together, but it is clear that not all robots use a system model. Â € Model œSensing € and â € € œActuation Models contain information with which to transform raw data into information-dependent physical task for the driver, and vice versa.
Â
Planning. This is the activity that predicts the outcome of possible actions and select the â € € œbestâ input. Almost by definition, planning can only be made on the basis of some kind of model.
Â
Regulation. This component processes the results of detection and planning components, to generate a reference point for action. Again, this activity regulation may or may not be based on some kind of (system) model.
Â
The term â € € œcontrolâ is often used instead of â € œregulationâ €, but it is impossible to clearly identify the domains that use one term or another. The sense used in the show's WEBook context.
Â
Balance on robotic systems
Â
This â € € œcomponentsâ description of a robotic system is complemented by a description œscaleâ € â €, ie, the scales of these systems have a great influence on the content specific planning, testing, modeling and control of components in a particular scale, and therefore also in the corresponding sections of the WEBook.
Â
Mechanical scale. The physical volume of the robot largely determines the limits of what can be done with it. Overall, a large-scale robot (for example, an autonomous container crane or a space shuttle) has different capabilities and problems of macro control of a robot (for example, a industrial robot arm), a desktop robot (such as â € € œsumoâ robots popular among fans), or milli micro or nano robots.
Scale space. There are big differences between robots operating in 1D, 2D, 3D, or 6D positions (three and three orientations).
Â
Timeline. There large differences between the robots must react within hours, seconds, milliseconds or microseconds.
Â
Scale of power density. A robot must be operated in order to advance, but actuators need space and energy, so the relationship between the two determines some capabilities of the robot.
Â
System complexity scale. The increasing complexity of a robot system with the number of interactions between independent subsystems, and components control must adapt to this complexity.
Â
Computational complexity scale. robot controller is inevitably runs on computer hardware real world, so are limited by the available number of the calculations, the communication bandwidth available, and the memory storage available.
Â
Obviously, these parameters are never applied independently scale the same system. For example, a system that must react in microseconds scale time scale can not mechanical macro or involve a large number of interactions with communication subsystems.
Â
Background sensitivity
Â
Finally, no description even scientific material is always entirely objective or context, in the sense that it is very difficult for taxpayers to WEBook € â € œforgetâ background during the writing of their contribution. In this respect, robotics has roughly twofold: (i) mathematics and engineering side, which is very â € € œstandardizedâ in the sense that there a broad consensus about the tools and theories for use (â € œsystems theory; €), and (ii) the face of AI, which is rather low standard, not for a lack of interest or research efforts, but due to the inherent complexity of behaviour.â € â € œintelligent Terminology and systems of thought from both sources are different, hence the WEBook accommodate sections of the same material but written from different perspectives. This is not a â € € œbugâ, but a â € € œfeatureâ: having different opinions in the context of the same WEBook can only lead to better understanding and mutual respect.
Â
Research in robotics engineering follows the bottom-up approach: working systems existing and spread and became more versatile. Robotics Research in artificial intelligence is top to bottom: on the assumption that a set of primitive low level is available, how can apply in order to raise € â € œintelligenceâ of a system. The border between the two approaches changes continually, as more and more â € € œintelligenceâ algorithmically thrown into the theoretical system. For example, the response of a robot sensor input was considered â € € œintelligent Behaviour late seventies and even eighties. Therefore, belonged to avian influenza Later showed that many of the sensor-based tasks such as monitoring the surface or visual tracking could be formulated as control problems with algorithmic solutions. Since then, they did not belong to the IA more.
Â
Â
Â
Technology Robotics
Â
Most industrial robots have at least the following five parts:
Â
Sensors, effectors, actuators, controllers and effectors commonly known as weapons.
Â
Many other robots also have Artificial Intelligence and effectors that help achieve mobility.
Â
This section discusses the core technologies of a robot. Click one of the links above or use the navigation menu at the right end.
Â
Robotics Technology – Sensors
Â
Most the robots of today are nearly deaf and blind. sensors can provide some limited information to the robot so you can make your job. Compared to the senses and abilities of even the simplest things of life, the robots have a long way to go.
Â
The sensor sends information in the form electronic signal back to cfontroller. sensors also control the robot information about its environment and let him know the exact position of the arm, or state of the world around him.
Sight, hearing, touch, taste and smell are the types of information we get from our robots can world. be designed and programmed for specific information that is beyond what our five senses can tell us. For example, a robot sensor can "see" in the dark, detect small amounts of radiation as invisible or movement is too small or fast for the human eye.
Â
Here are some of the sensors are used to things:
Â
Physical Property
 Technology
Â
BUMP contact switch
Ultrasonic Distance Radar, Infrared
Light level cell photos, cameras
Sound level microphones
Strain gauges Strain
Rotate encoder
Compasses Magnetism
Chemical odor
Temperature thermal infrared
Slope inclinometers, gyroscope
Pressure Gauges
Altimeters Altitude
Â
   The sensors can be simple and complex, depending on the amount of information needs to be stored. A switch is a simple on / off sensor used to rotate the robot and off. The human retina is a complex sensor that uses more than one hundred million light-sensitive elements (rods and cones).  sensors provide information to the brain robots, which can be treated in various ways. For example, we simply react to the sensor output, if the switch is open, if the switch is closed, go.Â
Â
Levels of Processing
Â
   To find out when the switch is open or closed, will have to measure the voltage going through the circuit, that is electronics. Now say you have a microphone and you want to recognize a separate voice and noise signal that is processing. Now you have a camera, and you want to take the picture pre-processed and now have to figure out what these objects are, perhaps, compared with a large library of drawings, that is computation. sensory data processing is a very complex thing to try to do, but the robot needs this to have a "brain."  The brain has to be capable of processing analog or digital cables to connect all support electronics to go with the team, and batteries to power the whole thing, in order to process sensory perception data. requires that the robot has sensors (energy and electronics), computers (more power and electronics, and connectors (for connecting all). Â
Â
Switch Sensors
Â
A switches are the simplest sensors all. work processing, to electronics (circuits) level. The overarching principle is that of a closed-open front circuit. If a switch is open, no current can flow, if it is closed, current can flow and be detected. This simple principle can (and is) used in a wide variety of ways.
Â
Change sensors can be used in a variety of ways:
Â
contact sensors: detect when the sensor is in contact with another object (for example, triggered when a robot hits a wall or take an object, which can even be whiskers)
Â
Limit sensors: detect when a mechanism has moved to the end of its range
Â
shaft encoder sensors: detect how often a shaft rotates to have the click of a switch (open / close) each time the shaft rotates (for example, triggers for each shift, thus providing rotations)
Â
  There are many points in common switches: push button switches, switches, mouse, key board keys, phone keys, and others. Depending on how a switch is connected, can be normally open or normally closed. Of course, this depends on your bot electronics, mechanics, and their simple but task. useful sensor for robot obstacle is a switch that says when it hit something, so you can back up and back. Even for a simple idea, there are many different ways of application.
Â
Light Sensors
Â
As the contact switches physical and light sensors measure the amount of light that affect a photodetector, which is basically a sensor. resistance The resistance of a photocell is low when lit, that is, when it is very lightweight, is high when dark. In this regard, a light sensor is actually a "dark" sensor. In creation of a photocell sensor, you will end up using the equations we have learned before, because you will have to cope with resistance ratio of the photo cell, and stress resistance in circuit. sensor electronics course and it be the construction of electronic and write the program to measure and use the light sensor output can always be manipulated to make it more simple and more intuitive. What surrounds a light sensor affects your properties. The sensor can bea protected and placed in several multiple ways. sensors can be arranged into useful configurations and isolated from each other with shields.
Â
Like switches, light sensors can be used in many different ways:
Â
light sensors can be measured:
Â
light intensity (such as light / dark s)
Â
differential current (difference between photocells)
Â
broken road (change / drop in intensity)
Â
light sensors can be protected and focused in different ways
Â
Its position and directionality of a robot can make a big difference and impact
Â
Polarized light
Â
"Normal" light emanating from a source non-polar, which means traveling to all guidance on the horizon. However, if there is a polarizing filter in front of a light source, only light waves of a certain orientation of the filter will through. This is useful because now you can manipulate this light was left with other filters, and if we put another filter with the same feature level, almost everything that is obtained of through. But if we use a filter perpendicular (one with a 90 degree angle relative property), will lock all the polarized light can be used light. to manufacture specialized photoelectric sensors simple, if you put a filter in front of a light source and the same or a different filter in front of the cell, which can be manipulated ably what and how much light detect.Â
Â
Resistive Position Sensors
Â
   We said earlier that a photocell is a resistive device. may also feel the resistance in response to other physical properties bending. such as the strength of the device increases with the amount that doubles bent. These sensors were originally developed for video game control (for example, Nintendo Powerglove), and usually very Note that repeated bending useful. sensor. will be spent is not surprising that a curve of the sensor is much less Robust light sensors, although using the same underlying principle of resistance.
Â
Potentiometers
Â
   These devices are very common for manual tuning, you've probably seen some controls (such as volume and tone sound equipment).  Commonly called pots, which allow the user to manually adjust the resistance. The general idea is that the device consists of a movable tap along ends. two sets as the tap moves, resistance changes. As you can imagine, the resistance between the two ends is fixed, but varies the resistance between the Mobile and end either as the party is moved. In robotics, the pots are commonly used to sense and position tuning for sliding and rotating mechanisms.
Â
Biological Analogues
Â
All sensors that we have described exist in the system biological
Â
Touch / contact sensors with much greater precision and complexity in all species
Â
Bend / receptor resistance musclesÂ
Â
Reflective optical sensors
Â
   We have mentioned that if we use a light bulb in combination with a photodetector, we can break a road sensor. This idea is the principle underlying reflective optical sensors: the sensor consists of a transmitter and a detector. Depending on the arrangement of the two to each other, we can obtain two types Sensor:
Â
reflectance sensors (emitter and detector are next to each other, separated by a barrier, the objects are detected when the light is reflected off them and back into the detector)
Â
road break sensors (emitter and detector front of other objects are detected if interrupts the light beam between the emitter and detector)
Â
   The transmitter is usually made out of a light emitting diode (LED) and the detector is usually a photodiode / phototransistor.
Â
   Note that these are not the same technology as photocells resistive. Resistive photocells are nice and simple, but its strength properties are slow photodiodes and photo transistors are much faster and therefore the type preferred technology.
Â
What can you do with this simple idea of the reflectivity of light? A whole bunch of useful things:
Â
object presence detection
Â
detection of distant objects
Â
surface feature detection (finding / following markers / tape)
Â
wall / ceiling tracking
Â
encoding the axis of rotation (with an encoder wheel with ridges or black and white color)
Â
decoding Barcode
Â
   Note, however, that the reflectivity of light depends on the color (and other properties) of a surface. A light surface reflects light better than dark, and a black surface can not reflect at all, with what appears invisible to a light sensor. Therefore, may be more difficult (less reliable) to detect dark objects in this way that the lightest. In the case of distance from the object, lighter objects are further away appear closer to the darkest objects that are not that far. This gives an idea of how the physical world is partially observable. Despite sensors that are useful, we have no complete and accurate information in full.
Â
   Another source of noise in light sensors is the ambient light. The best thing to do is subtract the ambient light sensor reading in order to detect real change in reflected light, no ambient light. How is that done? By taking two (or more, to be exact) readings of the detectors, one with the transmitter on, and with it off, and subtracting the two values together. The result is the level of ambient light, which can be subtracted from future readings. This process is called sensor calibration. Of course, remember that ambient light levels can change so that the sensors may need to be calibrated on several occasions.
Â
Road Break Sensors
Â
   We talked about the idea of breaking road sensors. In general, any pair of devices compatible with emitter-detector can be used to produce a sensor:
Â
an incandescent lamp and a photocell
Â
Red LEDs and photo transistors visible-light-sensitive
Â
or infrared emitters and IR detectors
Â
Axis coding
Â
encoder to measure shaft rotation angle of a shaft position mediation and / or speed information. For example, an indicator speed measures how fast the wheels of a vehicle is turning, while an odometer measures the number of rotations of the wheels.
Â
In order to detect partial or complete rotation, we must somehow make the decisive element. This is usually done by setting a round disc on the shaft, and cutting notches in it. A light emitter and detector are placed on each side of the disc, so that the notch passes between them, light passes, and is found where there is no notch on the disk, the light does not pass.
Â
If only one notch on the disk, and then a rotation is detected as is the case. This is not a very good idea, since it allows only one Low resolution for measuring speed: the smallest unit that can be measured is a complete rotation. In addition, some rotations may overlooked due to noise.
Â
In general, many notches are cut on the disc, and hit the light reaches the detector are counted. (You can see that it is important to have a speed sensor here, if the shaft rotates very quickly.)
Â
An alternative to reducing the notches on the hard disk is painted black (to absorb, reflect-) and white (highly reflective) spots, and measure the reflectance. In this case, the emitter and detector are on the same side of the disc.
Â
In either case, the sensor output will be a wave function of the intensity of light. This can be a process to produce the speed, counting the peaks of the waves.
Â
Note that the encoding axis measures both the position and the speed of rotation, by subtracting the difference in readings of position after each time interval. Speed, however, tells us how fast a robot moves, or if you're moving at all. There are several ways of using this measure:
Â
measure the speed of a powered (active) wheel
Â
person using a wheel that is dragged by the robot (measuring progress forward)
Â
We can combine position and velocity information for finer things:
Â
move in a straight line
Â
rotate by an exact amount
Â
Note, however, to do such things is very difficult, because the wheels tend to slip (noise effector and error) and slide and there is usually some settling and reaction gear mechanism. Shaft encoders can provide information to correct errors but with some error is inevitable.
Â
Quadrature Shaft Encoding
Â
Until Now, we talked about the position and speed detection, but not to speak of direction of rotation. Suppose that the wheel suddenly changes the direction of rotation, it would be useful for the robot detects that.
Â
An example of a common system to be measured position, velocity and direction is a computer mouse. Without a measure of direction, a mouse is pretty useless. How to measure the rotation?
Â
Quadrature Shaft Encoding is an elaboration of the basic idea break from the road, instead of using a single sensor, it takes two. The encoders are aligned so that the two flows Data from the detector and fourth cycle (90 degrees) out of phase, hence the name "square." When comparing the results of the two encoders at each step departure time of the passage of time, we can say if there is an address change. When both are included in the sample at each time step, only one of them will to change its state (ie, switching from on to off) at a time, because they are out of phase. What determines the direction of the axis is rotating. Every time a tree is moving in one direction, a counter is incremented, and when it becomes the opposite direction, the counter is decremented, thus keeping track global position.
Â
Other uses quadrature encoding axis robot arms with complex joints (the joints the swivel /-think your knee or shoulder), Cartesian robots (and printers large) where an arm / rack moves back and forth along a shaft / gear.
Â
Modulation and Demodulation of Light
Â
We mentioned that the ambient light is a problem because it interferes with the light emitted by a light sensor. One way around this problem is the emission of light modulated, ie, to quickly turn the transmitter on and off. This signal is much easier and more reliably detected by a demodulator, which is tuned to the particular frequency modulated light. Not surprisingly, a detector needs to detect several flashes in a row in order to detect a signal, ie to detect frequency. This is a small point, but is important in writing demodulator code.
Â
The idea of modulated IR light is commonly used, for example in household remote controls.
Â
Modulated light sensors are often more reliable than those based light sensors. They can be used for the same purpose: to detect the presence of an object to measure the distance to a nearby object (Requires intelligent electronics, see the course notes)
Â
Infra Red (IR) sensors
Â
Infrared sensors are a type of light sensors, which operate in the infrared part of the frequency spectrum. IR sensors are active sensors consist of: consists of a transmitter receiver. and infrared sensors are used in the same way that visible light sensors is that we have discussed so far: as balance beams and reflectance IR is sensors. preferable to visible light in robotics (and other) applications because it suffers a little less interference from the environment, as it can be easily modulated, and simply because is not visible.
Â
IR Communication
Â
modulated infrared can be used as a port serial for transmission of messages. This is a fact IR is how modems work. There are two basic methods:
Â
bit frames (including in the sample in the center of each bit, assuming all the bits to make the same amount of time for transmission)
Â
short intervals (more common commercial use, sampled on the trailing edge, the time interval between sampling determines whether a 0 or 1)
Â
Distance Ultrasonic detection
Â
As mentioned earlier, ultrasound screening is based on the principle of flight times. The emitter produces a sonar "chirp" sound, away from the source, and if barriers meetings is reflected in them and returns to the receiver (microphone). The amount of time it takes the sound beam to return is tracked (by starting a timer when the "chirp" is produced, and stop when the reflected sound returns) and is used to calculate the distance the sound travels. This is possible (and quite easy), because we know how fast sound travels, this is a constant, which varies slightly depending on the temperature.
Â
At room temperature, sound travels to 1.12 meters per millisecond. Another way to say that sound travels in 0.89 milliseconds per foot. This is useful to recall a constant.
Â
The process of finding a location based on sonar echolocation calls. The inspiration for the ultrasound screening comes from nature, bats use ultrasound instead of vision (this makes sense, living in dark caves where vision would be largely useless). Bat sonar is extremely sophisticated in comparison with artificial sonar, consisting in many different frequencies, used to find even the smallest quick flights prey and to avoid hundreds of other bats, and communication finding a partner.
                                                        Â
Specular Reflection
Â
A major disadvantage of ultrasound screening is its susceptibility to specular reflection (specular reflection means the outer surface of the object). While the sonar detection principle is based on the sound wave reflecting surface and returning to the receiver, it is important to remember that the sound wave will not necessarily be retrieved from the surface and "come back. "In fact, the direction of reflection depends on the incident sound beam angle and the surface. The smaller the angle, not the most higher the probability that the sound is limited to "shepherd" to the surface and bounce back again from the issuer, in turn generating a false long / distant reading. This is called specular reflection, as smooth surfaces with specular properties, tend to aggravate this problem reflection. Thick surfaces produce reflections irregular, some of which are more likely to return to sender. (For example, in our robotics lab on campus, using sonar sensors, and have aligned part of the test area with cardboard, because it has much better sound reflective properties of the very smooth wall behind him.)
Â
In summary, long sonar readings can be very inaccurate, because they can be false rather than accurate reflections. This should be taken into account when programming robots, or robot can produce highly undesirable and risky behavior. For example, a robot approaches a wall at a steep angle can not see the wall at all, and hit him!
Â
However, sonar sensors have been successfully used in highly sophisticated robotics applications, including mapping the terrain and cover, and continue a very popular choice in mobile robotic sensor.
Â
The first commercial ultrasonic sensor was produced by Polaroid, and is used to automatically measure the distance to the nearest object (presumably being photographed). These simple Polaroid sensors still remain the most popular sonar available on the market "(which come with a processor board that deals with analog electronics). Their level properties include:
Â
rank 32 feet
Â
30-degree beamwidth
Â
sensitivity to specular
Â
return shortest distance
Â
Polaroid sensors can be combined into phased arrays to create more sophisticated and more accurate sensors.
Â
One can find ultrasound used in a variety of other applications, the best known is that range from submarines. Sonars have focused much and have wider beams. Simpler and more mundane applications involve automated "tape-measures, measures of height, burglar alarm, etc.
Â
Machine Vision
Â
Until now, we've talked about relatively simple sensors. They were simple in terms of information processing that return. We now turn to machine vision, ie the cameras as sensors.
Â
Cameras, of course, the model biological eyes. Needless to say, all eyes biologics are more complex than any camera we know today, but as you can see, cameras and machine vision systems that process sensory information, not entirely simple! In fact, the vision is a difficult subject that has historically been a separate branch of Artificial Intelligence.
Â
The general principle of a camera is light, scattered from the objects in the environment (they are called there), passes through an opening ("Iris" in the simplest case a hole in the most sophisticated of a lens) and influence what is called flat image. In biological systems, the image plane is the retina, which is attached to numerous bars and cones (photosensitive elements) that in turn, bind to the nerves that carry out the "first look", and then passing the information through the brain to make "higher level "vision-processing. As mentioned earlier, a very high percentage of humans (and animals) of the brain devoted to visual processing, so this is a very complex task.
Â
In the chambers, instead of photosensitive rhodopsin and the rods and cones, we use silver halide photographic film or silicon circuits charge-coupled device (CCD) cameras. In all cases, some information about the incoming light (For such as intensity, color) is detected by these photosensitive elements in the image plane.
Â
In machine vision, the team must make sense of information obtained in the image plane. If the camera is very simple, and uses a small hole, then it requires some calculation calculate the projection of the objects in the environment in the image plane (note will be reversed). If it is a lens (as in the eyes of vertebrates and real cameras), then more light can enter, but the price of being focused, objects only a range of distances from the lens is in focus. This range distance is called depth of field camera.
Â
The image plane is usually divided into equal parts, called pixels, so generally arranged in a rectangular grid. In a typical camera has 512 by 512 pixels in the image plane (for comparison, there are 120 x 10 ^ 6 rods and 6 x 10 ^ 6 cones in the eye, willing hexagonal). Let's call the projection onto the image plane of the image.
Â
The brightness of each pixel image is proportional to the amount of light directed into the chamber through the revision of the object's surface that projects of that pixel. (This of course depends on the properties reflectance of the surface of the patch, the position and distribution of light sources in the environment, and the amount of light reflected by other objects in the scene patch surface.) As a result, the brightness of a patch depends on two types of reflexes, one of them is a mirror (not the surface, as we saw before), and the other is diffuse (light penetrates the object, is absorbed and then re-issued). To properly light reflection model and reconstruct the scene, all these properties are necessary.
Â
Suppose it is a black and white camera with 512 x 512 pixel image plane. Now we have an image, which is a collection of pixels, each intensity between white and black. To find an object in the image (if there is one, which of course we do not know a priori), the typical step first ("first window") is to detect edges, ie, find all the edges. How do you recognize? We define the edges as curves the image plane through which there is a significant change in brightness.
Â
A simple approach would be to observe the strong brightness changes by differentiation of the image and look for areas where the magnitude of the derivative is large. This almost works, but unfortunately it produces all sorts of spurious peaks, ie noise. Also, do not inherently able to distinguish the changes in intensity due to the shadows of those due to physical objects. But let's forget that for now and think about the noise. What about noise?
Â
We relaxed, ie, applies a mathematical procedure called convolution, which finds and eliminate isolated peaks. Convolution, in effect, applies a filter to image. In fact, to find arbitrary edges in the image, we must convolve the image with many filters with different orientations. Fortunately, relatively complicated mathematics involved in edge detection is well studied, and for now are not standard and preferred approaches of edge detection.
Â
Once we have the edge, the next thing to do is try to find objects from all edges. Segmentation is the process of division or organization of the image into parts that correspond to continuous objects. But how do we know that the lines correspond to objects, and what makes an object? There are several signs that can be used to detect objects:
Â
Models can be stored of line drawings of objects (from many angles, and in many different scales as possible!) and then compare these with all possible combinations of the edges of the image. Note that this is a very intensive and computationally expensive. This general approach, which has been extensively studied, is called model-based vision.
Â
We can take advantage of movement. If you look at a picture in two consecutive time steps, and moving the camera in the middle, each of the continuous solid objects (which obeys physical laws) will move as one, ie brightness properties will be retained. This hives us a clue to find objects by subtracting two images each other. But note that this also depends on knowing how the camera moved over the scene (direction, distance), and that nothing moved on the scene at that time. This general approach, which has also been extensively studied, is called vision of movement.
Â
We can use stereo (Ie, binocular stereopsis, two eyes / cameras / views). As with the vision of the movement, but without having to move in reality, we get two images, which may undermine each other if we know what the differences between them must be, that is, if we know how the two cameras are arranged / positioned relative to each other.
Â
We can use the texture. The patches are consistent with uniform texture and brightness are almost identical, so we can assume that from the same object. When removing can get a hint about which parts may belong to the same object in the scene.
Â
We may also use shading and contours in a similar manner. And there are many other methods, with the participation and projective invariant object shape, etc.
Â
Note that all the above strategies are used in biological vision. It is difficult to recognize objects or totally unexpected new (because we have no models at all, or not in the list.) The movement helps to get our attention. Stereo, ie, two eyes, is critical, and all carnivores use it (have two eyes pointing in the same direction, unlike the herbivores). The brain does an excellent job of quickly extracting the information we need for the scene.
Â
Machine vision has the same task of making real-time view. But this is, as we have seen, a very difficult task. Often, an alternative to trying to do all the steps above to make object recognition, it is possible to simplify the vision problem several ways:
Â
Use color, find specific items and unique colors, and recognizing that way (such as stop signs, for example)
Â
Use a small image plane, rather than a total of 512 x 512 pixel array, we can reduce our view at least, for example, only a line (called a linear CCD). Of course it is much less information than the image, but if we're smart, and know what to expect, we can process what we see quick and helpful.
Â
Use other, more simple and fast, sensors, and combine those with vision. For example, cameras IR isolate people by body temperature. Tweezers allows us to touch and move objects, after which we can ensure there.
Â
Use the information on the environment, and if you know they drive on the highway, which has white lines, specifically looking for lines in places correct the image. This is how the first and still the fastest road and highway driving is robotic.
Â
Those and many clever techniques must be employed when considering how important it is to "see" in real time. Consider Highway driving as a significant and increasing application of robotics and AI. Everything moves so fast that the system must perceive and act in time to react protective and secure, and intelligent.
Â
Now that you know how complex it is the vision, you can see why it was not used in the early robots, and still not used to all applications, and does not hesitate simple robots. A robot can be extremely helpful without the vision, but some of the tasks required. As always, is essential to devise a proper match between the sensors of the robot and the task.
Â
About the Author
Assistant professor in lord venkateswara engineering college.I am doing phd in sathyabama university, Tamil Nadu,India.
