Visual Event-Based Egocentric Human Action Recognition

Moreno-Rodríguez, Francisco J.; Traver, Vicente; Barranco, Francisco; Dimiccoli, Mariella; Pla, Filiberto

doi:10.1007/978-3-031-04881-4_32

Cited by 7 publications

(3 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Data Open Images juga digunakan pada riset domain lain seperti human action recognition [92], image retrieval [93], hingga visual question answering [94].…”

Section: Google's Open Imagesunclassified

Sintesis Teks Ke Gambar: Tinjauan Atas Dataset

Arifin

2024

EEICT

View full text Add to dashboard Cite

Penelitian ini bertujuan untuk melakukan analisis mendalam terhadap berbagai dataset yang digunakan dalam riset sintesis teks ke gambar. Fokus utama penelitian ini adalah pada pemahaman karakteristik masing-masing dataset, pengaruh pemilihan dataset terhadap hasil penelitian, serta keunggulan dan kelemahan setiap dataset. Beberapa dataset yang diteliti meliputi MS COCO, CUB-200-2011, dan Oxford 102 Flower, bersama dengan dataset-domain khusus lainnya yang relevan. Metode penelitian mencakup analisis deskriptif terhadap jumlah gambar, karakteristik visual, dan deskripsi teks yang melibatkan setiap dataset. Data yang diperoleh dianalisis secara kualitatif untuk mendapatkan wawasan mendalam tentang setiap dataset. Hasil analisis diharapkan dapat memberikan panduan bagi peneliti dalam memilih dataset yang sesuai dengan tujuan penelitian mereka dalam sintesis teks ke gambar. Penelitian ini diakhiri dengan rekomendasi dan kesimpulan yang merangkum temuan utama dan relevansinya dalam konteks riset ini.

show abstract

“…Data Open Images juga digunakan pada riset domain lain seperti human action recognition [92], image retrieval [93], hingga visual question answering [94].…”

Section: Google's Open Imagesunclassified

Sintesis Teks Ke Gambar: Tinjauan Atas Dataset

Arifin

2024

EEICT

View full text Add to dashboard Cite

show abstract

“…Also, event-based sensing has been employed to investigate action recognition on third-person view datasets, reflecting the increasing interest in this sensing paradigm [8]. In a recent study [9], the authors compared the performance of a CNN combined with a LSTM architecture on conventional gray-level frames with corresponding simulated event-based data with respect to human action recognition. Their results show the plausibility of using simulated event-based data to classify four different activities.…”

Section: Related Workmentioning

confidence: 99%

Fall Detection with Event-Based Data: A Case Study

Wang,

Risi,

Talavera

et al. 2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Fall detection systems are relevant in our aging society aiming to support efforts towards reducing the impact of accidental falls. However, current solutions lack the ability to combine low-power consumption, privacy protection, low latency response, and low payload. In this work, we address this gap through a comparative analysis of the trade-off between effectiveness and energy consumption by comparing a Recurrent Spiking Neural Network (RSNN) with a Long Short-Term Memory (LSTM) and a Convolutional Neural Network (CNN). By leveraging two pre-existing RGB datasets and an event-camera simulator, we generated event data by converting intensity frames into event streams. Thus, we could harness the salient features of event-based data and analyze their benefits when combined with RSNNs and LSTMs. The compared approaches are evaluated on two data sets collected from a single subject; one from a camera attached to the neck (N-data) and the other one attached to the waist (W-data). Each data set contains 469 video samples, of which 213 are four types of fall examples, and the rest are nine types of non-fall daily activities. Compared to the CNN, which operates on the high-resolution RGB frames, the RSNN requires 200× less trainable parameters. However, the CNN outperforms the RSNN by 23.7 and 17.1% points for W-and N-data, respectively. Compared to the LSTM, which operates on event-based input, the RSNN requires 5× less trainable parameters and 2000× less MAC operations while exhibiting a 1.9 and 8.7% points decrease in accuracy for W-and N-data, respectively. Overall, our results show that the event-based data preserves enough information to detect falls. Our work paves the way to the realization of high-energy efficient fall detection systems.

show abstract

“…Also, event-based sensing has been employed to investigate action recognition on third-person view datasets, reflecting the increasing interest in this sensing paradigm (Huang, 2021). In a recent study (Moreno-Rodríguez et al, 2022), the authors compared the performance of a CNN combined with a LSTM architecture on conventional gray-level frames with corresponding simulated event-based data with respect to human action recognition. Their results show the plausibility of using simulated event-based data to classify four different activities.…”

Section: Related Workmentioning

confidence: 99%

Fall detection by egocentric vision

Wang

View full text Add to dashboard Cite

IoT Internet of Things LBP Local Binary Pattern LSTM Long Short Term Memory LIF Leaky Integrate and Firing MLP Multi-layer Perceptron MACs Multiply and Accumulates NN Neural Network NNI Neural Network Intelligence RF Random Forest ROC Receiver Operating Characteristic RSNN Recurrent Spiking Neural Network SVM Support Vector Machine TP True Positives TN True Negatives Chapter 1capture data in diverse conditions, two cameras, namely RGB and infrared, were affixed to the subjects' neck and waist positions. This positioning facilitated data collection under varying illumination levels, both low and high, as well as in both indoor and outdoor environments. Consequently, this dataset lends itself to be utilized as a multi-modality resource, catering to both visual and audio-based analysis approaches. Thesis structureThe rest of the thesis is organized as follows. Chapter 2 provides a thorough review of the literature, describing the current state-of-the-art methodologies in fall detection. Chapter 3 elucidates a case study wherein a pre-trained CNN method is applied to data from a singular subject. In Chapter 4, we meticulously craft a late decision fusion methodology, amalgamating both traditional handcrafted and deep learning features. Chapter 5 introduces a streamlined Recurrent Spiking Neural Network, applied to transformed event-based data, and sets it in juxtaposition with a CNN + LSTM approach on the original RGB data. The thesis culminates in Chapter 6, where we present the summary, conclusions, and future perspectives.Additionally, the final part of the thesis includes the Appendix describing a new dataset, which is aimed at being used for benchmarking purposes.

show abstract

Visual Event-Based Egocentric Human Action Recognition

Cited by 7 publications

References 30 publications

Sintesis Teks Ke Gambar: Tinjauan Atas Dataset

Sintesis Teks Ke Gambar: Tinjauan Atas Dataset

Fall Detection with Event-Based Data: A Case Study

Fall detection by egocentric vision

Contact Info

Product

Resources

About