Action boundaries detection in a video

Abstract : In the video analysis domain, automatic detection of actions performed in a recorded video represents an important scientific and industrial challenge. This paper presents a new method to approximate the boundaries of actions performed by a person while interacting with his environment (such as moving objects). This method relies on a Codebook quantization method to analyze the rough evolution of each pixel and then decide whether this evolution corresponds to an action or not; this decision is taken by an automated system. Statistics are then produced - at the scale of the whole frame - to estimate the start and the end of an action. According to our proposed evaluation protocol, this method produces interesting results on both real and simulated videos. This statistic-based protocol is discussed at the end of this paper. The interpretation of this evaluation protocol nominates this method to be a solid base to localize the exact boundaries of actions or - in the framework of this research activity - to associate prescriptive text with a visual content.
Complete list of metadatas

Cited literature [21 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01445226
Contributor : Open Archive Toulouse Archive Ouverte (oatao) <>
Submitted on : Tuesday, January 24, 2017 - 4:27:32 PM
Last modification on : Thursday, October 17, 2019 - 8:54:22 AM
Long-term archiving on : Tuesday, April 25, 2017 - 6:24:52 PM

File

wehbe_17233.pdf
Files produced by the author(s)

Identifiers

Citation

Hassan Wehbe, Bassem Haidar, Philippe Joly. Action boundaries detection in a video. Multimedia Tools and Applications, Springer Verlag, 2015, vol. 75 (n° 14), pp. 8239-8266. ⟨10.1007/s11042-015-2748-5⟩. ⟨hal-01445226⟩

Share

Metrics

Record views

339

Files downloads

229