S PATIAL -C OLOR M EAN -S HIFT T RACKERS WITH N ORMALIZED F EATURE AND

and Weighted Information

Considering the background information, we add the weighted-background information to the trackers which we developed before. In this section, the tracker 1 is defined as (3-31) and the tracker 2 is defined as (3-32).

4.4.1 Face Sequence

‹ Tracker 1

Figure 4.4-1 : Face tracking results of spatial-color mean-shift tracker 1 with rg feature and weighted-background information in 3.6. Shown are frames 33, 93, 117, 126, 183, 256, 271, 455, 766. (red:

tracker 1 with rg feature and weighted-background information, blue: tracker 1 with rg feature only)

Figure 4.4-2 : Distance error of face sequence of spatial-color mean-shift tracker 1 with rg feature and weighted-background information in 3.6 that is compared with that with rg feature only. (*note: we only consider

the distance error which is smaller than 50 pixels)

Figure 4.4-3 : Iteration number of face sequence. (left: tracker 1 with rg feature and weighted-background information, right: tracker 2 with rg feature only)

Figure 4.4-2 shows that the performance of tracker 1 is more accurate than that of tracker 1 without the weighted-background information, and tracker 1 captures the target at all times. The average of iteration number of tracker 1 is larger than that of tracker 1 without weighted-background information, but the difference between these

two trackers is not large. In conclusion, the tracker 1 about the face tracking is much better.

‹ Tracker 2

Figure 4.4-4 : Face tracking results of spatial-color mean-shift tracker 2 with rg feature and weighted-background information in 3.6. Shown are frames 33, 93, 117, 126, 183, 256, 271, 455, 766. (red:

tracker 2 with rg feature and weighted-background information, blue: tracker 2 with rg feature only)

Figure 4.4-5 : Distance error of face tracking sequence of spatial-color mean-shift tracker 2 with rg feature and weighted-background information in 3.6 that is compared with that with rg feature only. (*note: we only consider

the distance error which is smaller than 50 pixels)

Figure 4.4-6 : Iteration numbers of face sequence. (left: tracker 1 with rg feature and weighted-background information, right: tracker 2 with rg feature only)

With the experiment result of tracker 2, Figure 4.4-5 also shows that the tracking results of tracker 2 is much better when adding the weighted-background information.

The location of tracking is more accurate, and the iteration number of tracker 2 is about the same as the tracker 2 without the weighted-background information as

shown in Figure 4.4-6.

‹ Tracker 1, Tracker 2, Traditional Mean-Shift Tracker

Figure 4.4-7 : Face tracking results of spatial-color mean-shift trackers with rg feature and weighted-background information in 3.6. Shown are frames 33, 93, 117, 126, 183, 256, 271, 455, 766. (red:

tracker 1 with rg feature and weighted-background information, blue: tracker 2 with rg feature and weighted-background information, green: traditional mean-shift tracker)

Figure 4.4-8 : Distance error of face sequence of spatial-color mean-shift trackers with rg feature and weighted-background information in 3.6. (*note: we only consider the distance error which is smaller than 50

pixels)

Figure 4.4-9 : Iteration number of face sequence. (left: tracker 1 with rg feature and weighted-background information, middle: tracker 2 with rg feature and weighted-background information, right: traditional

mean-shift tracker)

We can see that the tracker 1 is the best tracker from the real face sequence as shown in Figure 4.4-7. The distance error of tracking results of the tracker 1, tracker 2, and the traditional mean-shift tracker are compared in Figure 4.4-8. In these three trackers, the tracking locations of tracker 1 are the most accurate than those of tracker

2 and than those of traditional mean-shift tracker. The iteration numbers of three trackers are about the same. Summary of all, the spatial-color mean-shift tracker 1 is the best, and the spatial-color mean-shift tracker 2 is better than the traditional mean-shift tracker.

4.4.2 Cup Sequence

‹ Tracker 1

Figure 4.4-10 : Cup tracking results of spatial-color mean-shift tracker 1 with rg feature and

weighted-background information in 3.6. Shown are frames 4, 45, 63, 69, 81, 105, 166, 243, 364. (red: tracker 1 with rg feature and weighted-background information, blue: tracker 1 with rg feature only)

Figure 4.4-11 : Distance error of cup sequence of spatial-color mean-shift trackers with rg feature and weighted-background information in 3.6. (*note: we only consider the distance error which is smaller than 50

pixels)

Figure 4.4-12 : Iteration number of cup sequence. (left: tracker 1 with rg feature and weighted-background information, right: tracker 1 with rg feature only)

With this sequence of complex background and complex appearance of target, the performance of the tracker 1 is strongly improved, obviously. The tracker 1 captures the target at all times when the tracker 1 without weighted-background information

always loses the target.

‹ Tracker 2

Figure 4.4-13 : Cup tracking results of spatial-color mean-shift tracker 2 with rg feature and

weighted-background information in 3.6. Shown are frames 4, 45, 63, 69, 81, 105, 166, 243, 364. (red: tracker 2 with rg feature and weighted-background information, blue: tracker 2 with rg feature only)

Figure 4.4-14 : Distance error of cup sequence of spatial-color mean-shift tracker with rg feature and weighted-background information in 3.6. (*note: we only consider the distance error which is smaller than 50

pixels)

Figure 4.4-15 : Iteration numbers of cup sequence. (left: tracker 2 with rg feature and weighted-background information, right: tracker 2 with rg feature only)

Similar with tracker 2, the tracker is improved after adding weighted-background information and captures the target at all times.

‹ Tracker 1, Tracker 2, Traditional Mean-Shift Tracker

Figure 4.4-16 : Cup tracking results of spatial-color mean-shift trackers with rg feature and

weighted-background information in 3.6. Shown are frames 4, 45, 63, 69, 81, 105, 166, 243, 364. (red: tracker 1 with rg feature and weighted-background information, blue: tracker 2 with rg feature and weighted-background

information, green: traditional mean-shift tracker)

Figure 4.4-17 : Distance error of cup sequence of spatial-color mean-shift trackers with rg feature and

weighted-background information in 3.6. (*note: we only consider the distance error which is smaller than 50 pixels)

Figure 4.4-18 : Iteration number of cup sequence. (left: tracker 1 with rg feature and weighted-background information, middle: tracker 2 with rg feature and weighted-background information, right: traditional

mean-shift tracker)

As shown in Figure 4.4-16, the all trackers capture the target at all times, but the tracking locations of the tracker 1 and tracker 2 are more accurate than the traditional mean-shift tracker in the real image sequences. Figure 4.4-17 shows the distance errors of three trackers; the errors of tracker 1 and tracker 2 are about the same and are always smaller than those of traditional mean-shift tracker. The iteration numbers of the proposed trackers are quite similar with the traditional one as shown in Figure 4.4-18. To sum up, the trackers which we developed up to this point is already better than the traditional one.

4.4.3 Walking Girl Sequence

‹ Tracker 1

Figure 4.4-19 : Walking girl tracking results of spatial-color mean-shift tracker 1 with rg feature and weighted-background in 3.6. Shown are frames 28, 106, 111, 124, 130, 153, 166, 196, .220. (red: tracker 1 with

rg feature and weighted-background, blue: tracker 1 with rg feature only)

Figure 4.4-20 : Iteration numbers of walking girl sequence. (left: tracker 1 with rg feature and weighted-background, right: tracker 1 with rg feature only)

Under these circumstances of the variation of illumination and partial occlusion, Figure 4.4-19 shows that the tracker 1 captures the target when tracker 1 without weighted-background always fail. When the target has been covered by the cars from 106^th frame to 220^th frame, the tracker 1 still captures the target.

‹ Tracker 2

Figure 4.4-21 : Walking girl tracking results of spatial-color mean-shift tracker 2 with rg feature and weighted-background in 3.6. Shown are frames 28, 106, 111, 124, 130, 153, 166, 196, .220. (red: tracker 2 with

rg feature and weighted-background information, blue: tracker 2 with rg feature only)

Figure 4.4-22 : Iteration numbers of walking girl sequence. (left: tracker 2 with rg feature and weighted-background, right: tracker 2 with rg feature only)

Figure 4.4-21 shows that the tracker 2 fails during the tracking process. At 153^rd

frame, the tracker 2 fails and captures the background object because the appearance of the region which tracker 2 tracks is very similar with the girl model which we want to track. The tracker 2 uses the spatial information and the spatial information of the background region at 153^rd frame has very similar part, and this is the reason why tracker 2 fails.

‹ Tracker 1, Tracker 2, Traditional Mean-Shift Tracker

Figure 4.4-23 : Walking girl tracking results of spatial-color mean-shift trackers with rg feature and weighted-background in 3.6. Shown are frames 28, 106, 111, 124, 130, 153, 166, 196, .220. (red: tracker 1 with rg feature and weighted-background, blue: tracker 2 with rg feature and weighted-background, green: traditional

mean-shift tracker)

Figure 4.4-24 : Iteration numbers of walking girl sequence. (left: tracker 1 with rg feature and weighted-background, middle: tracker 2 with rg feature and weighted-background, right: traditional mean-shift

tracker)

All trackers are placed in Figure 4.4-23 to be compared the performance and results. To sum up, the tracker 1 always captures the target girl under the circumstances of the variation of illumination and partial occlusion, but the tracker 2 and traditional mean-shift fail in the tracking process. The spatial-color mean-shift tracker 1 is the best tracker which we developed.