A combined pose, object, and feature model for action understanding

Packer, Ben; Saenko, Kate; Koller, Daphne

doi:10.1109/cvpr.2012.6247824

Cited by 54 publications

(49 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The high-level model evaluated here is based on the joint activity recognition and object tracking method of [30], which presents a model for understanding the interactions between humans and objects while performing an action over time. The model uses an "in the hand" interaction primitive and represents a variety of actions in which an object is manipulated.…”

Section: High-level Modelmentioning

confidence: 99%

“…We combine this novel representation with two existing methods: a lowlevel latent state temporal sequence model [28], and a high-level model based on a sequence of structured features including primitive representations of object state and person-object interaction [30]. We provide a comprehensive evaluation of these representations, separately and in combination, on the VISINT corpus.…”

Section: Introductionmentioning

confidence: 99%

“…A recent model for recognition based on interaction primitives [30] was shown to offer strong performance on the VISINT corpus; this model exploited appearance and STIP primitives defined on person and object trajectories obtained using a deformable part model detector [11]. Such models are powerful, but still fail to track all types of objects, or be robust to many types of observation conditions.…”

Section: Introductionmentioning

confidence: 99%

“…Although a more intuitive framework, these representations suffer from the inherent inaccuracy of bounding box detection and from the fact that they do not model the entire scene and hence, cannot exploit contextual and geometric cues. On the other hand, the high-level approach adopted in our paper [30] can model activity The key-segments output (top row) automatically generates spacetime object segments that appear most central to the activity in the entire video clip. Unlike simple background subtraction (bottom row), they can distinguish the shapes of adjacent foreground objects, and extract object-like regions that do not move.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Mid-level Features Improve Recognition of Interactive Activities

Saenko¹,

Packer²,

Chen³

et al. 2012

Self Cite

View full text Add to dashboard Cite

We argue that mid-level representations can bridge the gap between existing low-level models, which are incapable of capturing the structure of interactive verbs, and contemporary high-level schemes, which rely on the output of potentially brittle intermediate detectors and trackers. We develop a novel descriptor based on generic object foreground segments; our representation forms a histogram-of-gradient representation that is grounded to the frame of detected key-segments. Importantly, our method does not require objects to be identified reliably in order to compute a robust representation. We evaluate an integrated system including novel key-segment activity descriptors on a large-scale video dataset containing 48 common verbs, for which we present a comprehensive evaluation protocol. Our results confirm that a descriptor defined on mid-level primitives, operating at a higher-level than local spatio-temporal features, but at a lower-level than trajectories of detected objects, can provide a substantial improvement relative to either alone or to their combination.

show abstract

Section: High-level Modelmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Mid-level Features Improve Recognition of Interactive Activities

Saenko¹,

Packer²,

Chen³

et al. 2012

Self Cite

View full text Add to dashboard Cite

show abstract

“…Packer et al [19] presented a system that is able to recognize difficult, fine-grained human actions with the management of objects in truthful action sequences. Reddy et al [20] propose the scene context info obtained from moving and immobile pixels in the key frames, in combination with motion features, to resolve the action recognition difficulty on a big dataset with videos from the web.…”

Section: Related Workmentioning

confidence: 99%

Support Vector Machine Based Approach for Translating Video Sceneries to Natural Language Descriptions

Wankhede¹,

Kagalkar²

2017

IJIVP

View full text Add to dashboard Cite

Human uses communication language either by written, spoken or typed to describe visual the world around them. So, the study of text description for any video goes increasing. This paper represents a framework that gives output as a description for any video having a maximum size of 50 seconds by using natural language processing. The framework is divided into two sections called training and testing. The training section is used to train the video with its description like activities of objects present in that video. The trained data is stored into the database with its features of scenario of video. Another section is testing section. The testing section is used to test the video and retrieve the output as description of video. By using Natural language processing sentences are generated from objects and their activities present in the video.

show abstract