A Cascade-Classifier Approach for Fall Detection

Putra, I Putu Edy Suardiyana; Brusey, James; Gaura, Elena

doi:10.4108/eai.14-10-2015.2261619

Cited by 5 publications

(34 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although Ojetola improved the detection rate, our prior work [ 14 ] showed that the method has a high computational cost because it has to extract features for every possible segment produced by the sliding window. A cascade-classifier approach (CCA) was developed to reduce this cost, and it also improved the classification performance.…”

Section: Related Workmentioning

confidence: 99%

“…Threshold-based approaches detect falls by checking if the measured acceleration exceeds a predefined (and fixed) value [ 6 , 7 , 8 ]. Machine learning-based approaches use labeled data to train a classifier using supervised machine learning algorithms (e.g., the support vector machine (SVM), decision tree, and/or artificial neural networks) that can recognize the characteristic features of falls [ 9 , 10 , 11 , 12 , 13 , 14 , 15 ].…”

Section: Introductionmentioning

confidence: 99%

“…Putra et al [ 14 ] proposed a new mechanism called the cascade-classifier approach (CCA), to reduce the computational cost of the system by extracting features only when the user’s body state is active. The body state is checked by using a 2-s FNSW and a threshold of 1.6 g (gravity).…”

Section: Introductionmentioning

confidence: 99%

“…The presence of multiple peaks (that we call the multi-peak problem) makes the estimation of the impact stage even harder as it can mislead the alignment of the beginning of that stage. Although the multi-peak problem is important for the data segmentation process, it has not been considered in previous studies [ 14 , 18 ].…”

Section: Introductionmentioning

confidence: 99%

“…For this application, reducing the computational cost of the algorithms running on the wearable device is important as it reduces power consumption and thus extends battery life [ 22 ]. In addition, we compared EvenT-ML to two existing fall detection approaches: a threshold-based approach (IMPACT+POSTURE) from Kangas et al [ 7 ] and a fall stages-based machine learning approach from Putra et al [ 14 ]. Compared to those techniques, our approach is able to achieve a significantly better F-score.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

An Event-Triggered Machine Learning Approach for Accelerometer-Based Fall Detection

Putra

Brusey

Gaura

et al. 2017

Sensors

Self Cite

View full text Add to dashboard Cite

The fixed-size non-overlapping sliding window (FNSW) and fixed-size overlapping sliding window (FOSW) approaches are the most commonly used data-segmentation techniques in machine learning-based fall detection using accelerometer sensors. However, these techniques do not segment by fall stages (pre-impact, impact, and post-impact) and thus useful information is lost, which may reduce the detection rate of the classifier. Aligning the segment with the fall stage is difficult, as the segment size varies. We propose an event-triggered machine learning (EvenT-ML) approach that aligns each fall stage so that the characteristic features of the fall stages are more easily recognized. To evaluate our approach, two publicly accessible datasets were used. Classification and regression tree (CART), k-nearest neighbor (k-NN), logistic regression (LR), and the support vector machine (SVM) were used to train the classifiers. EvenT-ML gives classifier F-scores of 98% for a chest-worn sensor and 92% for a waist-worn sensor, and significantly reduces the computational cost compared with the FNSW- and FOSW-based approaches, with reductions of up to 8-fold and 78-fold, respectively. EvenT-ML achieves a significantly better F-score than existing fall detection approaches. These results indicate that aligning feature segments with fall stages significantly increases the detection rate and reduces the computational cost.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

An Event-Triggered Machine Learning Approach for Accelerometer-Based Fall Detection

Putra

Brusey

Gaura

et al. 2017

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

A Machine Learning Approach for Fall Detection and Daily Living Activity Recognition

2019

View full text Add to dashboard Cite

The number of older people in western countries is constantly increasing. Most of them prefer to live independently and are susceptible to fall incidents. Falls often lead to serious or even fatal injuries which are the leading cause of death for elderlies. To address this problem, it is essential to develop robust fall detection systems. In this context, we develop a machine learning framework for fall detection and daily living activity recognition. We use acceleration and angular velocity data from two public databases to recognize seven different activities, including falls and activities of daily living. From the acceleration and angular velocity data, we extract time-and frequency-domain features and provide them to a classification algorithm. In this paper, we test the performance of four algorithms for classifying human activities. These algorithms are the artificial neural network (ANN), K -nearest neighbors (KNN), quadratic support vector machine (QSVM), and ensemble bagged tree (EBT). New features that improve the performance of the classifier are extracted from the power spectral density of the acceleration. In the first step, only the acceleration data are used for activity recognition. Our results reveal that the KNN, ANN, QSVM, and EBT algorithms could achieve overall accuracy of 81.2%, 87.8%, 93.2%, and 94.1%, respectively. The accuracy of fall detection reaches 97.2% and 99.1% without any false alarms for the QSVM and EBT algorithms, respectively. In a second step, we extract features from the autocorrelation function and the power spectral density of both the acceleration and the angular velocity data, which improves the classification accuracy. By using the proposed features, we could achieve overall accuracy of 85.8%, 91.8%, 96.1%, and 97.7% for the KNN, ANN, QSVM, and EBT algorithms, respectively. The accuracy of fall detection reaches 100% for both the QSVM and EBT algorithms without any false alarm, which is the best achievable performance.

show abstract

A Machine Learning Approach for Fall Detection Based on the Instantaneous Doppler Frequency

Chelli

Pätzold

2019

IEEE Access

View full text Add to dashboard Cite

Modern societies are facing an ageing problem that is accompanied by increasing healthcare costs. A major share of this ever-increasing cost is due to fall-related injuries, which urges the development of fall detection systems. In this context, this paper paves the way for the development of radio-frequencybased fall detection systems, which do not require the user to wear any device and can detect falls without compromising the user's privacy. For the design of such systems, we present an activity simulator that generates the complex path gain of indoor channels in the presence of one person performing three different activities: slow fall, fast fall, and walking. We have developed a machine learning framework for activity recognition based on the complex path gain. Additionally, we propose a novel method that accurately estimates the instantaneous Doppler frequency (IDF) from the complex path gain. Then, we extract six features from the IDF and provide the feature vector as input to the classifier, which has to predict the user's activity. We assess the recognition accuracy of four different classification algorithms: K-nearest neighbors (KNN), decision tree, artificial neural network (ANN), and cubic support vector machine (SVM). Our analysis reveals that the KNN, decision tree, ANN, and cubic SVM achieve an overall recognition accuracy of 86.1%, 94%, 98.9%, and 99.9%, respectively. The best performing algorithm, cubic SVM, has a fall detection accuracy of 100% with zero false alarms and zero undetected falls, which represents the best achievable performance. By comparing our fall detection system with existing ones in the literature, we demonstrate the superiority of our proposed solution.

show abstract

A Cascade-Classifier Approach for Fall Detection

Cited by 5 publications

References 18 publications

An Event-Triggered Machine Learning Approach for Accelerometer-Based Fall Detection

An Event-Triggered Machine Learning Approach for Accelerometer-Based Fall Detection

A Machine Learning Approach for Fall Detection and Daily Living Activity Recognition

A Machine Learning Approach for Fall Detection Based on the Instantaneous Doppler Frequency

Contact Info

Product

Resources

About