| Summary: | 碩士 === 國立交通大學 === 資訊學院資訊學程 === 106 === Safety driving has been intensively studied recently. Existing work can be classified into three directions. The first one is to assist driving by Advanced Driver Assistance System (ADAS). The second way is to detect driver’s phone and forbid driver from using it. The last
one is to use driver gesture recognition to avoid driver taking their eyes off the roads.
In this paper, we propose Ultrasound Driver Gesture Detection System (UDGDS). The main idea is to detect driver’s gestures. We use a cell phone to detect driver’s gestures. First, we transmit 20kHz acoustic signal from speaker, and sense the Doppler effect of reflected signals from the dual microphones on cell phone. Then user motions are detected. If there is a motion detected, we use Gabor filter to extract some features from received signals, and use Support Vector Machine (SVM) classifier to classify hand gesture. Last, we decide whether the gesture is made by driver or passenger.
Our work can be realized directly by off-the-shelf cellular phones. We conduct our experiments at lab and in vehicle. In these two environments, we collect 6 gestures each with 100 times to test our gesture recognition performance. We reach accuracy beyond 95% . These 6 gestures are each tested 30 times for distinguishing driver from passenger and we reach gesture classification accuracy of 90% and 83% accuracy for distinguishing driver and passenger,respectively.
|
|---|
