computer vision refers to the ability of computers to see, and to make sense of objects they see just as humans and animals do.
I am not going to talk about the methods used today to achieve computer vision. Those can be obtained online with ease.
Rather, I am going to identify the shortcomings of these methods(both ones announced by the industry and also ones I have personally observed).
I will later provide solutions(algorithms) I think would improve upon current methods. So that we can achieve computer vision.
LIST OF SOME(BUT NOT ALL) ISSUES FACING COMPUTER VISION METHODS TODAY
They require a huge number of pictures to train models.
The pictures used to train the models all have to be of specific sizes and shapes or you have to provide each of them in as many different sizes and shapes as possible.
3.The pictures require labelling(naming). You have to name all of them so that the computer can easily identify what the image refers to.
4.The mathematics used in computer vision involve decimals, fractions and square roots. They also make extensive use of statistics and probability.
Therefore, the computer is never 100% sure of its answer.
5.If a part of an object in the input image is hidden or not clear enough, for example, a bicycle with one tire hidden behind a wall because the bicycle was parked against the wall, the computer cannot identify this object as a bicycle.
- They use models.
"All models are approximations. Essentially, all models are wrong, but some are useful".
- George Box
Top comments (0)