How it all began
To teach a computer to “see”, you first need a sensor that reads
Printed text proved to be a perfect teststand for the first generation of algorithms - it was enough to increase the contrast of the image and find transitions from white to black and vice versa. This made it possible to identify the boundaries of the letters and their shape, after which it was possible to determine each letter by comparing it with previously prepared samples.
Building open source libraries such asOpenCV has opened the door to experimenting with arbitrary images. By the shape of the identified edges of the object, the computer could understand where a person, a human eye, a bird, a car were depicted - computer vision systems went beyond the boundaries of text recognition.
Security cameras appeared, giving a signalif a person appeared in the frame. Applications with computer vision have settled in mobile phones: identification by face, fingerprint, retina or QR code scanner have long been common features. And now, not a primitive pattern comparison algorithm is responsible for understanding what is depicted, but a neural network trained using machine learning technology.
Why computer vision is better than human
The first and obvious advantage of CV is muchwider range of perception. Computer vision, unlike ours, is not limited to the visible spectrum. The main thing is that the sensor is able to create an image for further processing, and what is the source of this image - a thermal imager, lidar, video camera, X-ray machine, ultrasound or MRI machine - does not matter.
The second advantage of computer vision systems islack of a human factor. They do not get tired, are not distracted, do not want to sleep, they have a 360-degree view. The necessary information can be requested from the archive for re-viewing.
In addition, in the learning process, the neural networkcan reveal connections that are not obvious to human perception. For example, the system analyzes images of human internal organs and forms an idea of how they should look. So it is possible to determine not only the type of organ, but also the degree of deviation from the norm - in particular, to detect a tumor that is not diagnosed by other methods.
How computer vision is changing the way we view the world
Computer vision technology nowThey are used in medicine for the detection of cancerous tumors, in industry for assembly control at the production stage and in cars as additional security systems. We do not hesitate to use computer vision to unlock the device, and we habitually swear at cameras on the roads that issue fines.
Computer vision is changing even the perception of sports- automatic tracking systems control broadcast cameras, help to sort out controversial moments - whether the ball remained within the tennis court or whether it crossed the goal line, whether there was a violation or an offside position.
Records are kept with the help of computer visionthe congestion of vehicles. The system monitors how many rear wheels of the truck are in contact with the road surface, how much the tires are deformed, and calculates the vehicle's payload. All that remains is to recognize the license plate and make an entry in the database.
Computer vision is irreplaceable in warehouseaccounting. It works like this: the scanner reads the barcode on the box with the product and enters the information into the database. Moreover, it is not necessary to scan every box - it is much more convenient to read QR codes from all boxes on the pallet at once.
Processing systems continue to evolvetexts - in modern document management systems, a scan of a contract or invoice is downloaded as a PDF document, but thanks to computer vision, it can be searched for, classified or copied.
A separate area is the integration of computervision and space technology. In agriculture, the state of the crop is monitored using satellites, the Ministry of Emergency Situations monitors the occurrence of fires and floods. Google has used satellite imagery to use computer vision to identify and map all structures in Africa - from dugouts dug in a hill on the river bank to high-rise buildings.
Computer vision technologies are needed forcreation of augmented reality - it is thanks to the recognition of reference points that it is possible to correctly embed the synthesized image into the frame and combine real and virtual spaces. For example, in an application for remote jewelry fitting, a smartphone camera reads the position of the user's hand, builds a model and adds the selected ring to it. The already familiar 3D masks on Instagram are also one of the manifestations of computer vision in combination with VR / AR.
Gaining popularity and highly specializedapplications that give out information about a tree from a photograph of a leaf help mushroom pickers find out if they got an edible trophy, and collectors can quickly get information about a coin by simply pointing a smartphone camera at it.
What are the disadvantages of computer vision systems
Computer vision systems are extremely sensitiveto hardware resources. The higher the resolution of external sensors and the more computing power, the more reliable and accurate the recognition process. This can be seen in the example of face recognition technologies for unlocking mobile phones. A decade ago, this required good lighting and a well-defined distance from the face. Today the phone recognizes the owner even at night.
Another CV disadvantage that can be attributed to"Growing pains" - errors in the classification of similar objects. For example, a person can easily distinguish a monkey from a kangaroo, and a lizard from a crocodile, but for a computer it is not so easy with similar poses and certain shooting angles.
Another limitation of computer vision isdecision making process. A modern premium car can see farther than a person: it has no blind spots, does not get in the way of darkness at night or fog, it can detect the approach of an object by its thermal radiation. But in a real situation on the road, a person is still better at making decisions taking into account the current situation. And in order to increase safety, the autopilot has to monitor not only the traffic situation, but also its own driver.
In the world of machine vision, identification is one of thebasic processes, which immediately raises many questions related to privacy and personal data. Therefore, it is necessary to take into account the legal requirements and ethical restrictions associated with the very fact of a computer's assessment of human actions and the possible consequences of decisions made by algorithms. The topic of mistakenly written fines is a clear confirmation of this.
What the future holds for us
One of the most common horror storiesconnected with computer vision - the image of a "digital concentration camp", total control from which it is impossible to hide. Computer vision systems of the future will be able to control not only movements, but also intentions: frowning means sad, looking around means planning something. Fantastic dystopian blockbusters also add phobias and negativity in anticipation of such futures.
Screenwriters love to tickle their nerves, scaring new onestechnologies. In practice, the world today is "transparent" for control technologies. But the advantages of the introduction of computer vision will allow in the future not to worry about many routine things. For example, we will most likely forget about cash registers in supermarkets - the system itself identifies you at the entrance, controls what you put in the basket, calculates the cost at the exit and debits the funds from the account itself. Thanks to computer vision, it will be possible to quickly redirect traffic flows in order to avoid traffic jams, and traffic in general will become safer.
Computer vision is the technology of the futurewhich is already affecting our lives. It is deeply integrated into the modern world and we use its capabilities every day. But we are still at the very beginning of the path, where we have to not only solve the technical issues of increasing computing power, but also learn to live in a new reality. To do this, you will have to defeat phobias and come up with legislation that will prevent technology from getting out of control.
Read more:
Look at pictures of two merging galaxies, taken 9 years apart
Elon Musk: Starship can lift 1,000 times more cargo than other rockets
Dark matter reverses ordinary particles and fills the universe