Driver Activity And Gesture Recognition Using Fine-grained WiFi Signals

The driver's monitoring system may provide vital information about the state of driver attentiveness,vehicle maneuvering,and controlling.Besides the traditional approaches that rely on camera or wearable devices,wireless technologies for in-vehicle activity and gesture monitoring have emerged with remarkable attention.The existing recognition systems require dedicated devices for satisfactory recognition performance,which results in high cost.The ubiquity of fine-grained channel state information(CSI)of WiFi signals motivated us to accomplish an adequate framework for device-free in-vehicle activity and gesture recognition,utilizing off-the-shelf WiFi devices.The proposed low-cost solution requires neither sophisticated camera support to capture images nor any special hardware to carry with the entity.In this dissertation,we have developed and implemented three different models for the WiFi CSI-based driver's activity and gesture recognition.The first model is concerned with an efficient radio-image feature extraction method for driver's inattention monitoring.The second model is relevant to the driver's activity recognition using a multi-layer classification model.In the third model,we proposed an integrated classification algorithm for driver gesture recognition.The main contributions are summarized as follows:(1)In an activity recognition system,the efficient feature representation may reduce the burden of a classification algorithm.Recently,WiFi CSI-based radio-image descriptors have emerged with better recognition results.The major drawback of image features is computational complexity,which increases exponentially,with the growth of irrelevant information in an image.It is still unresolved how to choose appropriate radio-image features to alleviate the expensive computational burden.This research work exploits radio-image processing to address the problem of driver's activity monitoring with efficient computation of image features.To achieve this goal,we firstly adopt a cumulative moving variance to detect the presence of activity,and then the discriminant components are selected by analyzing the Eigen vectors.Next,the image features,i.e.gray level co-occurrence matrix(GLCM)and Gabor wavelet features,are extracted from the discriminant components of CSI-transformed radio-images.Finally,the extracted features are further analyzed using stacked sparse auto-encoder(S-SAE)in a semi-supervised manner to get more optimized features.The experimental results show that the proposed scheme is computationally efficient with less execution time as compared to traditional methods.The presented model could achieve a recognition accuracy of 93.1%to characterize the attentive and inattentive status of a driver.(2)The existing methods for human activity recognition rely on single-layer classification algorithms that do not guarantee maximum performance in multi-class classification problem.In this dissertation,a multi-layer classification model is provided to achieve higher scalability as compared to the traditional single-layer framework.In this novel approach,multiple classifiers are arranged in a hierarchy for better recognition performance leveraging the CSI of WiFi signals.This research study investigates the capability of the sparse least square support vector machine(SLS-SVM)with Baye's maximum likelihood estimation(MLE)for better discrimination of different activity-classes.Firstly,the activity detection is performed using the variance of the received signal.Afterward,the activity profile data is extracted,and then principal component analysis(PCA)is used to reduce the dimensions.Next,we extract the most representative statistical feature.Finally,SLS-SVM with Baye's likelihood estimation is used to recognize different activities.This sophisticated scheme leads to address the problem of multi-class classification for a complete description of in-vehicle activities including driving maneuvers,driver distraction,and fatigue recognition.The multi-layer model increases the overall performance by utilizing different feature sets and multiple classifiers for better adaptation to various applications.The simulation results show that the proposed ubiquitous model leads to improve the recognition performance significantly with an average recognition accuracy of 91.5%and a less execution time.(3)This research work provides an efficient WiFi CSI-based gesture recognition system to control secondary functions in a vehicle without diverting the driver's visual attention from the road.The conventional device-free indoor gesture recognition methods are not suitable for an in-vehicle cluttered environment.In this study,we demonstrate a novel classification model by integrating the sparse representation classification(SRC)and a variant of the K-nearest neighbors(KNN)algorithm.Both KNN and SRC algorithms have been efficiently used in various CSI-based device-free activity/gesture recognition systems.Conventional SRC is efficient in the sense that all coefficients participate well in decision making,however,it is time consuming that a testing sample is usually represented by all training samples.Hence,the computational burden of the SRC algorithm increases with the increase in the size of training data.In this typical approach,we have explored the performance of the mean of nearest neighbors(MNN)to address the problem of the expensive computational cost of SRC.In particular,we have used the decision rule of SRC to supervise MNN,and sparse representation coefficients are computed from the mean vector of nearest neighbors.Firstly,we estimate the K nearest neighbors from all training samples of each class and calculate the mean of K nearest neighbors within each class.Afterward,the decision rule of SRC is used to supervise MNN,and sparse representation coefficients are computed from the mean vector of nearest neighbors.Finally,the class of testing sample is decided on the residual between the testing sample and the mean of nearest neighbors within each class.The proposed WiFi CSI-based system may lead to overcoming the problem of driver's visual-manual distraction and can recognize human gestures very accurately for the application of the in-vehicle infotainment system.Experimental results show that the provided prototype outperforms the traditional methods with an average recognition accuracy of 90.6%in promising application scenarios.To ensure the reliability of the obtained results,different evaluation metrics,i.e.recognition accuracy and execution time are adopted.From the experimental investigations of the suggested models,it is concluded that the efficient feature extraction method and classification algorithm play a vital role to significantly improve the WiFi CSI-based activity/gesture recognition accuracy.To sum up,this dissertation has provided novel schemes to realize in-vehicle activity/gesture recognition with better performance in terms of pervasiveness,low cost,high accuracy,and less execution time.