Vision-based Vibration Detection Algorithm And Its Implementation

Under strong radiation environment, for the purpose of protecting staffs, it’s necessary to use specific mechanical instruments to replace human beings to work. Manipulator used under radiation environment is an example. Because of the requirements of operation space or task, sometimes the manipulator needs to be long and thin and it will be easy to vibrate too. When under normal environment, there are many ways to measure the displacement and vibration introduced by material flexibility. While it will not be the same situation,when under extreme environment; for example, when under strong radiation environment,traditional measurement sensor will introduce severe error or even lose it functionality.The measurement system based upon vision will be more accurate and more capable of resisting disturbance. Based on the above background, The main contents this thesis talking about is the vibration detection algorithm and its implementation.In order to get interframe displacement sequence, we combine FAST feature and KLT algorithm and then implement a faster algorithm. The commonly used methods that are used to get displacement sequence from image sequence including background differencing, interframe difference method, Mean-Shift algorithm and KLT algorithm.When under actual application scenario, Camera is fitted with manipulator end and move with it, this does not meet the condition that background differencing and interframe difference method require camera to be static. In addition, the image that the camera can capture does not contain separable colorful region, which limits the use of Mean-Shift algorithm. As a result, we choose KLT algorithm last. While KLT algorithm has its own shortage: KLT algorithm requires huge floating point arithmetic. This defect will cost huge calculation resource and make KLT algorithm unavailable under some platform that is lack in computing power. For the purpose of speeding up the execution of KLT algorithm as much as possible, we use FAST Feature to replace KLT feature which is tracked by tracking part of KLT algorithm. At the time of implementing this algorithm, in order to make full use of the hardware platform features, this thesis also talks about how to use SSEinstructions to accelerate the tracking part of KLT algorithm.The verification under Windows system shows that modifications mentioned above can speed up the execution of KLT algorithm effectively and get desired result. At last, we transplant the modified KLT algorithm into QNX and do some tests. The tests show that the modifications we do can speed up the execution of KLT algorithm under QNX effectively too.Image acquiring is the fundament of the work mentioned above. USB camera is low in price and widely used, so we choose to use it as the image acquiring device. Because QNX didn’t support UVC devices, we choose to develop UVC device driver according to UVC standard.