Detector runs a Harris/Fast9 Corner implementation on the GPU to detect features. For Harris Corner implementation, detector adds new features to the end of the given list of feature points. For example, if the maximum feature count is 2000 and the list has 1500 features, currently only up to 500 features will be added, starting from index 1500 onwards. The decision if a new feature is added is dependent on proximity to features already in the list as well as a threshold on the corner response function. For Fast9 Corner implementation, detector adds detected features to the start of given list of feature points while users are free to decide how to combine newly detected features with previously detected features.
Descriptor computes encodings for each of the detected feature points using the pixel information. For ORB descriptor implementation, the module computes the binary descriptor for detected features by comparing the intensity values of pixels. These are subsequently matched using Hamming distance.