Energy-Based Periodicity Mining With Deep Features for Action Repetition Counting in Unconstrained Videos

Yin, Jianqin; Wu, Yanchun; Chao-ran, Zhu; Yina, Zijin; Liu, Huaping; Dang, Yonghao; Liub, Zhiyi; Liuc, Jun

doi:10.1109/tcsvt.2021.3055220

Cited by 7 publications

(5 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The work of Yin et al [12] shares some similarities with our work, as it also extracts periodic features from a video with a learning-based method, reduces it to a 1D signal, and counts the repetitions with an algorithm relying on the Fourier transform. However, their approach is not generalizable to other types of data since it uses a neural network that is pre-trained on a large annotated video dataset (Kinetics [13]) in a supervised way.…”

Section: Related Workmentioning

confidence: 89%

“…As most of these methods ( [9,11,10,12,30]) are trained on a human motion video dataset (Countix being built on top of Kinetics), they are well adapted to human gestures and actions. However, this makes them (i) specific to videos and not any other type of input data and (ii) biased towards human motion.…”

Section: Related Workmentioning

confidence: 99%

“…This issue is often circumvented by pre-training such networks on large annotated datasets [5,6], but then the model may be biased towards specific visual features which may not be relevant for the task at hand and thus lead to a lower performance [7,8]. To tackle the periodicity counting problem in videos, state-of-the-art methods [9,10,11,12] are trained on Kinetics [13], a videos dataset of persons doing repetitive actions. Such pre-trained models are thus domain specific to human gestures, and their performance are likely to drop when used on less frequent domains, such as astrophysics or medical videos.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Periodicity counting in videos with unsupervised learning of cyclic embeddings

Jacquelin

Vuillemot

Duffner

2022

Pattern Recognition Letters

View full text Add to dashboard Cite

Section: Related Workmentioning

confidence: 89%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Periodicity counting in videos with unsupervised learning of cyclic embeddings

Jacquelin

Vuillemot

Duffner

2022

Pattern Recognition Letters

View full text Add to dashboard Cite

“…Time-Frequency Analysis: Tom F. H. Runia [5] utilized flow-based wavelet transforms instead of Fourier transforms to handle nonstationary and non-steady repetitions, and collected 100 instances of non-stationary cyclic repetitions in the QUVA Reptition dataset. Yin et al [6] employed high-energy rules and Fourier transforms to extract periodicity from videos and implemented repetitive counting using peak detection.…”

Section: Repetitive Action Countmentioning

confidence: 99%

Repetition Count: Application to Aero-engine Blade Counting Tasks

Kaiyu,

Huitao,

Jun

et al. 2024

Preprint

View full text Add to dashboard Cite

Engine blades, being critical components of aircraft engines, pose a substantial threat to both the engine and the entire aircraft if they fracture during flight. Hence, inspecting and maintaining these blades are crucial to ensuring flight safety. In the process of blade damage detection, personnel typically utilize borescope inspection equipment to manually examine each blade and count them as they pass, thereby guaranteeing the examination of every individual blade within the engine to prevent any missed or duplicate inspections. This paper presents a new video interpretation method applied to the scenario of engine blade counting. The core of this algorithm involves employing the cosine correlation function to calculate the similarity between video frames captured during borescope inspections, followed by adaptively thresholding the processed signal for dynamic binarization, and ultimately counting the falling edges. By adopting frame-related approaches instead of relying on local image characteristics, this algorithm exhibits high robustness against smooth blade surfaces and metallic reflections. Additionally, it efficiently manages motion blur and directional variations that occur during the rapid movement of the blades. Compared to existing methods, this algorithm requires minimal training time, is compatible with various turbine engine blades, and guarantees real-time count updates.

show abstract

“…9-Energy-Based Algorithms. They use the energy of the pixels in an image to detect movement [21]. 10-Feature-Based Algorithms.…”

Section: Figure 1 Rack Distribution In An Equipment Sitementioning

confidence: 99%

Motion detection using a Raspberry Pi 4 and OpenCV

Luna¹,

Rangel²,

Aceves³

et al. 2023

IJLEMR

View full text Add to dashboard Cite

Currently, one of the most important parts of the cloud and computing platforms are the data centers, since they house the equipment where the applications and information processing are executed. The vitalkey areas of data centers are the equipment sites. In addition to the fact that access is controlled and restricted to these areas, continuous monitoring of the presence of people in these areas is critical. The data centers have closed circuit television systems that allow this activity to be carried out; however, these systems do not send alert messages that allow timely reaction to any threat from unauthorized persons who have accessed the equipment site. Similarly, presence detectors of different technologies are used, which present disadvantages compared to the system presented here. For these reasons, in this work a motion detection system was developed based on the processing of video images captured with a camera integrated into a Raspberry Pi 4 card and programmed with Python and OpenCV functions. When the system detects movement, it sends a WhatsApp alert message to a mobile phone. The tests carried out showed that the system is accurate to 98.08%.

show abstract

Energy-Based Periodicity Mining With Deep Features for Action Repetition Counting in Unconstrained Videos

Cited by 7 publications

References 27 publications

Periodicity counting in videos with unsupervised learning of cyclic embeddings

Periodicity counting in videos with unsupervised learning of cyclic embeddings

Repetition Count: Application to Aero-engine Blade Counting Tasks

Motion detection using a Raspberry Pi 4 and OpenCV

Contact Info

Product

Resources

About