opencv-cpp | repository consists of a bunch of OpenCV programs | Computer Vision library

Here's a widget to select static ROIs from a video frame. Essentially the idea is to use cv2.setMouseCallback() and event handlers to detect if the mouse has been clicked or released. For this implementation, you can extract coordinates by holding down the left mouse button and dragging to select the desired ROI. You can reset the ROI using the right mouse button. To use the widget, press c to pause the video and start cropping. Then you are free to select the ROI. Once you have selected your ROI, press c again to crop the desired section. To resume the video, press r.

First of, resizing is a bad idea. Unless both images are resized together (which won't solve the problem), resizing will change the final result in undesired ways (e.g. the foreground object will appear larger than intended).

Alpha blending is usually used to add foreground elements to a background image. Therefore, I would fix the size of the background image, and consider it also the output image size. In applications, one might need to have the foreground exit the background image, but that is a specific application which requires more input (how to extend the background borders?).

Since the background image has a fixed size, we need a way to handle the alpha blending of a smaller image. Consider the simplified case, where the smaller (foreground) image is aligned with the larger (background) image at the point (0,0). Then, you can iterate over the background image, check if it overlaps the foreground image, and if it does, blend them.

Solving the general cases introduces another problem: positioning. You need to know where to place the foreground element. This requires some additional input.

Given a smaller image and a position where you want to place it, you can alpha blend against a larger image using an algorithm as follows:

Those tracking algorithms are for tracking complex patterns, like a face in a video where the lighting changes, and the direction of the head is changing. If in you application you can use color or light to mark your object uniquely, then that will be the fastest algorithm and simplest. if you need very precise localization, then you need to consider using a marker, the reason is that a marker allows sub-pixel localization.

You were almost there, except for two small mistakes.

The first mistake is using cv2.imread to load the HDR image without specifying any flags. Unless you call it with IMREAD_ANYDEPTH, the data will be downscaled to 8-bit and you lose all that high dynamic range.

When you do specify IMREAD_ANYDEPTH, the image will be loaded as 32bit floating point format. This would normally have intensities in range [0.0, 1.0], but due to being HDR, the values exceed 1.0 (in this particular case they go up to about 22). This means that you won't be able to visualize it (in a useful way) by simply casting the data to np.uint8. You could perhaps normalize it first into the nominal range, or use the scale and clip method... whatever you find appropriate. Since the early visualization is not relevant to the outcome, I'll skip it.

The second issue is trivial. You correctly scale and clip the tone-mapped image back to np.uint8, but then you never use it.

Script

The following code was used with cv2 version of 3.3.1.

I closely followed the opencv docs and it worked fine.

You cannot send opencv Mat inside ros topic system since Image topic has another coding system.

You need to use cv_bridge to convert images that you got from Opencv to ROS-image Format

there are some details and examples Here

To work with picamera, I recommend you to use the package picamera. You can follow this tutorial to deploy picamera: https://www.pyimagesearch.com/2015/03/30/accessing-the-raspberry-pi-camera-with-opencv-and-python/