Since there are distinctive color differences between the target and the background, the CBWH and LBWH models locate the target well except in the case where the ball touches the bat in Frame 26.
From the second row of Table 1, as far as target localization accuracy is concerned, the joint CB-LBWH method performs slightly worse than CBWH. LBWH performs best since the foreground features of the target model are well exploited in target foreground change scenes.
Because of the changing background and heavy occlusion during tracking, the target and background are hard to distinguish by taking account of only background color features, thus CBWH locks the target early in the period but loses it before the end.
After many consecutive occlusion frames, since the feature confidence is ignored by CBWH, the CBWH tracker will tend to track the similar object close to the target.
When the ball moves quickly, the CBWH and LBWH methods fail to track the target and then gradually recover the correct tracking.
Compared with CBWH and LBWH, the joint CB-LBWH model needs the lowest average number of iterations.
Subsequently the CBWH and LBWH schemes are combined and the resulting scheme is designated the appellation: joint CB-LBWH.