1
$\begingroup$

I am so much curious about how do we see(with eyes ofc) and detect things and their location so quick. Is the reason that we have huge gigantic network in our brain and we are trained since birth to till now and still training. basically I am saying , are we trained on more data and huge network? is that the reason? or what if there's a pattern for about how do we see and detect object. please help me out, maybe my thinking is in wrong direction. what I wanna achieve is an AI to detect object in picture in human ways.thanks.

$\endgroup$
  • $\begingroup$ Thanks for posting.Add some references also for effective feedback. $\endgroup$ – quintumnia Oct 25 '18 at 17:50
2
$\begingroup$

Object detection can conceivably imitate what the human visual system does. Research along these lines began in the 1980s in multiple laboratories and was termed Computer Vision.

How do we see ... and detect things and their location so quick[?] Is the reason that we have huge gigantic network in our brain and we are trained since birth to till now and still training[?]

Neurological visual systems continue to train as long as the eyes and neurons continue to function. A person may, long after doing well in sports, learn how to catch glass objects that fall off a counter using peripheral vision, using the same neural pathways as were used for visually coordinated movement in sports but. They are newly trained to the new cognitive intention of saving glassware and clean-up time.

The mapping of signals from the rods and cones to the visual cortex has been studied and some generalities have influenced artificial vision research and development. The convolution kernels used in CNNs are partly derived from biological research. In summary, the way mammals and other organisms with the equivalents of retinas see is via a set of transformations and pattern matching circuits. The sequence of information types through biological visual networks can be summarized.

  • Light as a function of compound angle and time.
  • Edges in both space and time
  • Elements of objects and actions — These are real world features, not in the machine learning sense but in the natural language use of the term features.
  • Objects and actions

The development of object and action recognition in computers follows this same basic sequence. The goal stated in the question is this.

Achieve an AI to detect object in picture in human ways

Something like this has been the goal of many government, academic, and corporate laboratories, however the ability to recognize objects in still pictures is not as valuable as detecting the trajectories and other expressions of objects (such as changes in facial expressions) in real time.

Automated aiming, piloting, walking, and driving have been computer vision objectives since the first control systems were developed for anti-aircraft defense systems. Considerable effort is being invested into how steering, breaking, and signaling can be automated in road vehicles. A large array of household and industrial devices are being developed that rely on mapping the environment and navigating a robotic device through the map.

Stationary object recognition work is largely surrounding either categorizing images or drawing vector graphics from raster images, but these functions have limited application. As mobile devices develop further and images on the web give way to videos on the web, a very strong and uninterrupted trend, the temporal dimension will continue to gain importance in the field computer vision.

None of these objectives are novel. They are all based on animal capabilities, only some of which are uniquely human, and have been the subject of research since before the advent of digital computers. However the feasibility of many designs has improved, largely due to a significant reduction in costs for fixed computing resources. The vision system hardware of today costs roughly one thousandth of its 1990 cost.

$\endgroup$
0
$\begingroup$

If you have studied about Convolution Neural Network, you probably know how present day object detection algorithms works. Basically, an Artificial Neural Network tries to mimic the way our brain might work. So, CNN is probably the way our brain works to perform object detection, trying to recognize each small parts at each stage, that slowly grows to compile into the larger object to be recognised. However since, we are not restricted to data from just 2D plane, so the processing might be more complicated. We all learn things after being trained to do so. You could never have recognized a particular fruit if you have never seen it. The way we store the information, we gain, is different, that helps us to remember/recognise objects after only a few look. Until, we have more research that provides more details into the working of our brain, we are all good with Artificial Neural Network. Honestly, we are still far away from achieving general artificial intelligence. Now, if you want to implement an object detection algorithm, CNN is your only way.

$\endgroup$
  • $\begingroup$ Thanks for answering,can you also add some references to make it great. $\endgroup$ – quintumnia Oct 25 '18 at 17:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.