Abstract | ||
---|---|---|
Human activity prediction is a quite challenging problem, for activity prediction systems have to make decisions based on partially observed videos. However, activity prediction methodologies have many practical applications in real-world scenarios, such as human-robot interaction, security surveillance, etc. In this paper, we formulate the activity prediction problem into a discriminative weighted voting framework by unifying two different discriminative weights: 1) detector weights (DW): Mid-level features detected by distinct detectors are supposed to have different weights. 2) temporal weights (TW): Human activities are composed of sequential frames, so features in different phases should have different voting weights. We evaluate our voting model on two datasets. The experimental results clearly show that our method with discriminative weights enables better prediction of human activities. |
Year | DOI | Venue |
---|---|---|
2015 | 10.1145/2808492.2808494 | ICIMCS |
Field | DocType | Citations |
Voting,Pattern recognition,Computer science,Weighted voting,Artificial intelligence,Discriminative model,Machine learning | Conference | 0 |
PageRank | References | Authors |
0.34 | 10 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Zhen Xu | 1 | 23 | 4.21 |
Laiyun Qing | 2 | 337 | 24.66 |
Jun Miao | 3 | 220 | 22.17 |