The PIV raw images were used for flame detection and therefore have the same optical structure. Only the relevant parts of the cycles from the ignition point onwards were processed in Matlab. The raw images with a flame were also masked and the local standard deviation calculated. This local standard deviation was normalized using the standard deviation of an image without a flame. Areas of vaporized oil droplets differ in the standard deviation from areas with oil droplets and the associated high local intensity fluctuation. These differences in the values can be used to identify the area of burnt gas, referred to below as the flame, using a threshold value and to binarize the PIV images. The binarized images were averaged over the recorded cycles and normalized to one, indicating the probability that the flame reached this pixel at a given time. Statistical analyses were carried out on the basis of such flame probabilities.