6-5 Haar features 2

You also need to take into account the problem of one step in the traversal process. This template slides 10 pixels at a time, so a total of 9 slides are required. In addition to the first No. 0 time, the total in the horizontal and vertical direction respectively have 10 such templates, a total of 100 templates. To calculate these 100 templates, the Haar feature on the whole picture can be calculated. And if our step size is set to 5*5, then this computation increases by four times times. In fact, just such a slide is not over, and for each template there are several sections to scale to calculate the end. For example, the 10*10 may have to be scaled 1 times and 2 times, increasing by 10% each time.

And if we want to process the image in real time, then the image processed per second can not be less than 15 frames per second.

It takes 100 billion times to finish calculating all the Haar features on a picture. After the calculation we also have to the haar characteristics of the decision, such as Haar features commonly used in face recognition or target detection, so the computational volume is very large.

The calculation method based on the integral graph can greatly reduce the computational amount. However, even in this case, in some high-precision images and low performance embedded platform, the operation of this slider is still very large.

6-5 Haar features 2