Tracking a target in a raw image may not provide the best tracking performance in all cases. Often tracking a target in a processed image is better. The pre-processing may be edge detection such as a Sobel filter (shown here), or setting a threshold, noise filtering etc.
Using the Set Memory control the original image is saved to memory. The pre-processing (convolution) is performed on the main pixels. The target is tracked in the processed image, but the crop that is passed to the next control is extracted from the image in memory (if it exists). The type of pre-processing will vary depending on the video used.
The latest version (2.3.2) released on this web site (not in the MS store yet) has a new “Low Frequencies Center” option added to the FFT filter control. This option is more convenient for common FFT filters.
By including the low frequencies in the center of the FFT mask window a low pass filter can be implemented.
By excluding the same low frequencies a high pass filter can be implemented.
Keeping the resolution low (256×256 in this case) will provide a reasonable live experience; still not quite real-time.