globally-optimal greedy algorithms for tracking a variable number of objects
DESCRIPTION
Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects. Hamed Pirsiavash, Deva Ramanan , Charless Fowlkes Department of Computer Science, UC Irvine. Estimate number of tracks and their extent Do not initialize manually Estimate birth and death of each track. - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/1.jpg)
Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects
Hamed Pirsiavash, Deva Ramanan, Charless Fowlkes
Department of Computer Science, UC Irvine
![Page 2: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/2.jpg)
2
Estimate number of tracks and their extent– Do not initialize manually– Estimate birth and death of each track
![Page 3: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/3.jpg)
3
Our approach: Graph theoretic problem
– Globally Optimal • for a common class of objective functions
– Locally Greedy • and hence straightforward to implement
– Scale linearly in the number of objects and video length
![Page 4: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/4.jpg)
4
• Object state
• Object track
• K-object tracker
• Discretize state space S (e.g., scanning window locations)• Assume no tracks overlap (for now)• Must infer K, track births & deaths, and solve data association
![Page 5: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/5.jpg)
5
Trellis graph
• Local cost of window• Pairwise cost of transition• Dynamic programming finds a single track– (Viterbi algorithm)
![Page 6: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/6.jpg)
6
Trellis graph
• Local cost of window• Pairwise cost of transition• Dynamic programming finds a single track– (Viterbi algorithm)
Add edges to model occlusion
![Page 7: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/7.jpg)
7
What about variable length tracks?
How to find more than one track?
Birth cost
Death cost
![Page 8: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/8.jpg)
8
Equivalent graph problem: Min-cost-flow A generalization of min-cut/max-flow problem
Input flow of d
Output flow of d
A detection window
A transition
Introduced in “Zhang, Li, Nevatia, CVPR’08”
![Page 9: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/9.jpg)
9
Our contributionFind 4-track solution given a 3-track solution
DP
![Page 10: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/10.jpg)
10
Find 4-track solution given a 3-track solution
DPSub-optimum
Optimum
Our contribution
![Page 11: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/11.jpg)
11
Our contributionFind 4-track solution given a 3-track solution
DPSub-optimum
Optimum
SSP
Shortest path: New track can “suck” flow from existing tracks
![Page 12: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/12.jpg)
12
Solutions
• Globally optimum– Previous work
• Zhang et al CVPR’08: Introduced the model with a naïve solver
– Our algorithm• Exploits the special structure of graph (DAG, unit-capacity)• Is greedy using successive shortest path
• Approximate – Dynamic programming
• Is greedy
![Page 13: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/13.jpg)
13
Why is greedy nice?
Non-max-suppression in the loop:– At each iteration, suppress all windows overlapping with
the instanced track.
One iteration
![Page 14: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/14.jpg)
14
Experiments
Datasets:– Caltech pedestrian dataset
• Camera on a moving car• ~120,000 frames
– ETHMS dataset• Moving camera on a cross walk• ~1000 frames
![Page 15: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/15.jpg)
15
Detection vs false positive per frame (FFPI) for ETHMS dataset
![Page 16: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/16.jpg)
16
Detection vs false positive per frame (FFPI) for ETHMS dataset
![Page 17: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/17.jpg)
17
Detection vs false positive per frame (FFPI) for ETHMS dataset
![Page 18: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/18.jpg)
18
Novel, scalable, greedy algorithm with state-of-the-art results
![Page 19: Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects](https://reader035.vdocuments.us/reader035/viewer/2022062301/56816134550346895dd08ca8/html5/thumbnails/19.jpg)
Thanks