基於紅外線陣列感測器之居家老人定位與跌倒偵測系統

A Fall Detection System Based on Infrared Array Sensors with Tracking Capability for the Elderly at Home

陳維漢(Wei-Han Chen)

指導教授 ： 馬席彬

國立清華大學/電機資訊學院/電機工程學系/碩士(2014年)

摘要

近年來，隨著老人援助需求的增長，健康照護系統的開發漸漸地朝向具備舒適度、隱私及智慧化的方向發展，並以提升老年人的生活品質為宗旨。在本論文中，基於隱私問題的考量，我們採用低解析度的紅外線陣列感測器做為健康照護系統的開發，此紅外線感測器是由 16 × 4 的熱電堆陣列感測器所組成，可視角為 60˚ × 16.4˚，在陣列感測器之中，每個像素可取得所測得的溫度資訊以產生紅外線溫度影像，最後利用溫度影像實現定位及跌倒偵測的應用。
在系統之中，我們使用兩顆紅外線陣列感測器，布置在房間不同位置的牆上，以不同角度擷取人體在空間上的資訊。在前處理的部分，利用人體與環境溫度差異的特性，人體前景的影像可藉由將紅外線影像與背景溫度模型相減而得到，此背景模型會隨著背景溫度的變化更新。在定位方面，我們利用前景區域的溫度，計算出人體對應到每個感測器的方位角(AOA)，並藉由AOA定位法估測出人體位置，接著經由回歸模型修正估測位置的誤差，最後，我們的定位演算法可達到13.39公分的平均距離誤差。
除了定位之外，我們利用感測器擷取跌倒動作的特徵，實現跌倒偵測的演算法，在系統中，兩個感測器會同時捕捉人體的動作，我們會選擇前景面積較大的感測器資料去做特徵擷取的動作，其中擷取的特徵包含了7種特徵，最後利用這些特徵建立k-近鄰演算法(k-NN)的模型，進行跌倒偵測的分類。為了評估跌倒偵測的分類效果，我們使用80個跌倒動作及80個正常動作，並以10次交叉驗證(10-fold cross-validation)來選擇最佳的特徵集合與k-NN模型的k值，最後的分類結果達到95.25%的敏感度，90.75%的特異度及93%的準確度。

關鍵字

Infrared ； fall detection ； tracking

並列摘要

As the need of the elderly assistance increases, the healthcare monitoring systems are designed for improving the quality of life. The development strategy is aimed at comfortability, privacy, and intelligence. In this thesis, a low resolution privacy preserved infrared array sensor are adopted. The sensor is composed of a 16×4 thermopile array with the corresponding 60˚×16.4˚ field of view. To capture the infrared image, each pixel of infrared sensor contains the temperature value seen by the sensor. By using the infrared array sensors, our healthcare monitoring system is developed for the applications of tracking and fall detection.
In our system, two infrared sensors are attached to the wall at different places and are used for capturing the three dimensional image information. Before the tracking process, the foreground of human body is determined by subtracting the image with the background model using the temperature difference characteristic. The background model would update with the background temperature adaptively. Using the temperature value within the foreground region, the angle of arrival (AOA) from each sensor is obtained. The location is then estimated by the AOA based positioning algorithm. The estimated position is passed to the regression model to reduce the positioning error. As a result, the mean error of our tracking algorithm is 13.39 cm.
In addition to the tracking application, the fall detection algorithm is implemented by extracting the features from the falling action. Two sensors would capture the action at the same time. The sensor with larger foreground region is chosen for the feature extraction process, in which 7 features are included. These extracted features are applied to the k-nearest neighbor (k-NN) classification model for the fall detection. To build the k-NN model, 80 fall actions and 80 normal actions are collected from five subjects for estimating the performance. Using the 10-fold cross-validation method, the k value of k-NN model and the feature selection are determined by scanning the k value and every possible combination of feature subsets. Finally, 95.25% sensitivity, 90.75% specificity and 93% accuracy are achieved in our fall detection system.

參考文獻

[3] W.-T. Tang, C.-M. Hu, and C.-Y. Hsu, “A mobile phone based homecare management

long-distance health care network using minimized portable sensors and active alert sys-

tem,” in IEEE International Conference on Network-Based Information Systems (NBiS),

sit-stand transitions using a miniature gyroscope and its application in fall risk evaluation

in the elderly,” IEEE Transactions on Biomedical Engineering, vol. 49, no. 8, pp. 843–