improve jpeg size estimate

[jepler]

The jpeg size estimate in capture_buffer_size is causing folks trouble:

        if self.colorspace == OV5640_COLOR_JPEG:
            return self.width * self.height // self.quality

replace it with a better heuristic after gathering information.

note to users affected by this: Instead of using capture_buffer_size in jpeg mode, manually determine the size for your quality & resolution settings. When this value is too small, the capture routine returns None in jpeg mode.

[jepler]

espressif's esp32-camera uses width * height / 5 regardless of quality level.

[jmangum]

Confirming that setting the size of the capture buffer to cam.capture_buffer_size does for many settings of cam.quality truncate the buf = bytearray(cam.capture_buffer_size) such that the end-of-image (EOI) bit is missing, and cam.capture(buf) returns None rather than a memoryview jpeg. As noted, key is to set the size of buf large enough. After quite a bit of experimentation, where I am capturing images of a wide range of scenery, so far it appears that using esperssif's esp32-camera value of cam.width*cam.height/5 (perhaps this is supposed to be a floor divide, or //5?) works well. Should also note that I am using a cam.quality = 10, as I found that it reduces capture jpeg file sizes by about a factor-of-two while maintaining good image quality. Will deploy my remote camera system with this setting and see how it goes.

[jepler]

(the quoted code was "C" code in which "/" behaves similar to Python floor division "//" when the arguments are of integer type)

[jmangum]

Makes sense that the correct python usage should be floor division (//).

[jerryneedell]

FYI -- cam.width * cam.height // 5 seems to work much better for me as well.