Techniques for calibrating a video detection system are disclosed. A videoconferencing system may include a sensor configured to capture sequential frames of video image data during a videoconference. A processing subsystem may be configured to, using a multidimensional filter, generate data indicative of active areas of interest detected within respective ones of a plurality of video frames captured by the sensor, wherein the multidimensional filter is configured to identify active areas of interest (AAOIs) based at least on part upon a programmable density threshold. The processing subsystem may further: determine, based on the data indicative of AAOIs, that the videoconferencing system is in an unstable state; perform a calibration routine to identify one or more threshold values that reduce system instability; and apply the one or more threshold values to the multidimensional filter during further generation of data indicative of active areas of interest.