Demo Video of Tracking Results based on Partial Trajectory Grouping
|
Introduction:
|
||
| Results on the Grouped Tracklets, and Tracked Objects (Click to View Larger Images) | ||
| a. Temporal Segment from Frame 1500-1700 | ||
|
||
|
|
|
| Video results of our methods for the above segment | Video results of one prev methods for the above segment | |
| b. Temporal Segment from Frame 2700-2900 | ||
|
||
|
|
|
| Video results of our methods for the above segment | Video results of one prev methods for the above segment | |
| c. Temporal Segment from Frame 3500-3700 | ||
|
||
|
|
|
| Video results of our methods for the above segment | Video results of one prev methods for the above segment | |
| Some Quantitative Results for Performance Comparison | ||
|
||
| In the above figure, we compare the rate of positive object detection of our method to Ref.[2], which is computed from the whole 8-minutes long video. Given an acceptance threshold, we do an association of tracked objects with the ground-truth objects, and compute the positive detection rates. Owing to its capability to detect deformable objects with various sizes, we found that the proposed method has a higher rate of positive detections when the acceptable bias goes higher than 220mm, which is a tolerable bias for general surveillance applications. Furthermore, our method includes the tracking functionality intrinsically, which is another advantage over previous methods where an external tracking approach is needed to link their detected results, e.g., [1] and [2]. | ||
|
[1] F. Fleuret, J. Berclaz, R. Lengagne, and P. Fua, “Multicamera people tracking with a probabilistic occupancy map,” TPAMI, vol. 30, no. 2, pp. 267–82, 2008. [2] D. Delannay, N. Danhier, and C. De Vleeschouwer, "Detection and recognition of sports(wo)men from multiple views," in ICDSC 2009, Sept.2009, pp.1-7. |
||
| Other Supplemental Results | ||
| 1.Results on Shadow Removal | ||
| Conventional background extraction usually cannot separate shadows from other moving object. Here, we first improve the background extraction by inserting a shadow detector to remove the shadows from the detected forground occupancy masks. | ||
|
||
| Source Video Frames | ||
|
||
| Foreground Occupancy Masks from Conventional GMM based Background Extractor | ||
|
||
| Detected Shadow Area (in Pure White Color) | ||
|
||
| Refined Foreground Occupancy Masks with Shadow Removed | ||
| 2.Results on Ghost Removal | ||
|
||
| Figure on Ghost-removed Homography Plane | ||
| 3.Results on Extracted Partial Trajectories | ||
|
||
| Video results of extracted partial trajectories | ||
Maintained by chen-fan AT jaist.ac.jp. Last update 2012-01-20
