基于被动红外传感器的室内人行为机器学习模型

Machine Learning Model of Indoor Occupant Behavior Based on Passive Infrared Sensor

摘要:

室内人行为的准确识别，包括人员位置和活动类型的判定，是智能家居领域中各类电器设备实现多场景控制模式的重要输入参数。采用被动红外（PIR）传感器阵列监测人行为，分析人员不同位置及不同强度动作的数据特征。基于机器学习算法建立室内人员位置及动作识别模型，并对比不同累加时长和机器学习算法的模型预测准确度。最终以PIR传感器当前1 min的计数累加值（分钟计数值）及其前30 min计数累加值作为模型输入，选取随机森林算法构建了位置及动作识别模型。该模型在训练数据集十折交叉验证下准确率为99.9%，对新测试数据集的预测准确率为88.3%，能够识别实际人员的活动位置和动作强弱，具有一定的有效性和通用性。

Abstract:

Accurate recognition of indoor occupant behavior， including the recognition of position and activity type， is an important input for multi-scene control mode of various electrical equipment at intelligent homes. In the study described in this paper， the passive infrared （PIR） sensor array is used to monitor indoor occupant behavior. After analyzing the data characteristics of different positions and different activity intensities， based on the machine learning algorithm， the indoor occupant position and activity recognition model is established and the recognition accuracy of different cumulative time and machine learning algorithms are compared. The cumulative count value of this minute and of the previous 30 minutes of the PIR sensors are selected as the model input and the random forest algorithm is used to construct the final position and activity recognition model. The accuracy of the model is 99.9% under the 10-fold cross-validation of the training data set， while 88.3% under the new test data set， which shows that the position and activity recognition model has a certain validity and generality.