Research On Passive Sensing Techniques Based On Smart Devices

In recent years,the number and variety of smart devices are increasing explosively.Sensors equipped on smart devices are more plentiful and their functions are already per-fect.However,simple sensing applications can not satisfy users any more as users' ex-pectations becoming higher.Smart sensing-based applications is becoming a trend.As a kind of smart sensing technologhy,passive sensing techniques stand out because of their low cost and ubiquitousness.While how to apply passive sensing techniques into smart devieces still remains to be explored.So,in this paper,we do some research on passive sensing techniques based on smart devices.The first research is smartphone-based dangerous vehicle detection.As the number of traffic accidents caused by smartphone addicts is increasing every year,we propose a smartphone-based mechanism to help smartphone addicts to be aware of the approaching vehicle.we use the microphone to collect the voice data and then analyze the acoustic wave to detect the approaching vehicle.We propose a series of time-frequency domain features to differeniate the audio induced by a moving car from the background noise.Experimental results show that,this mechanism is robust in various environments.Even in a noisy one-way street,the recognition rate is still as high as 90%.The second research is gesture perception based on Wi-Fi signal.Nowadays,keyboard-based HCI is already out of date.Gesture recognition is becoming increasingly important.Researchers have implemented gesture-based HCI using camera and sensors.But cam-era will not work at night and sensor-based HCI requires additional devices on users.In the paper,we propose Wi-Fi signal-based gesture perception.We make CSI as our data source,and model the relationship between CSI and gesture,and propose a gesture percep-tion mechanism using statistical methods.Then we discover that amplitude distribution is an effective feature to differentiate various gestures and train a classifier using SVM.We design five gestures and three scenes and collect hundreds samples to evaluate the model.Result shows that the accuracy stays above 80%.