Study on the Automatic Basketball Shooting System Based on the Background Subtraction Method

Table 2

Methods used to update background models.

Method name	Description	Characteristics
Median	Median value between continuous multiple-frame sequences as the grayscale value of pixel in fixed time scales	Inconveniently obtains time sequence
Mean	As to the “median” method but the average value of frame sequence as the pixel value	Sensitive to light variation in the environment and dynamic background
Kalman filter	Predicts image transform results on the basis of Kalman filter theory	Long time to eliminate noise and uncontrollable procedure process
Single Gauss	Takes each grayscale value of pixel as the stochastic variable, and the whole process follows Gauss distribution	Convenient calculation process, but bad performance in a complicated scenario
Multiple Gauss	Superimposes a single Gauss process, multimodal situation in a complicated scenario	More model to superimpose and complies with a complicated scenario
Nuclear density	A nonparameter method estimates the current pixel value in a certain moment using nuclear density function	Prior distribution is not needed to know before calculating the density function of the sample