Image Transmitting And Object Positioning In Remote Vision Control System

With the popularity of internet and wireless network, and improvements of computer processing, remote vision control systems become a research hotspot in recent years. The work of this thesis is based on a C/S mode remote vision control system of a robot arm. The user could manipulate the robot arm through remote handling. The moving images of robot arm are collected by a camera installed on the local control sever. The images or video sequence are sent into the local computer and then analysed and compressed, which yield image location data and compressed image data. The image location data are then converted to physical locations and transformed through Internet to the remote client together with the compressed image data. The client accepts these data, then decompresses the images and displays them which makes the monitoring site available to the user. at the same time, the remote computer will send back control commands according to the location data.In the section of image compression, we use respectively DCT(discrete consine transform), EZW(embedded zerotree wavelet) and hybrid predicting/DCT combined with wavelet transform methods. 1) The storage method of the quantized DCT coefficients accelerates coding speed. And, an adaptive neighborhood blocking filter is added in the decoder to eliminate the blocking effect. 2) In the process of digital image compression based on wavelet transfom, the embedded zerotree wavelet (EZW) could achieve progressively transmiting of image data. For the problem of high space and time complexity that exited in EZW, we proposed three improvement methods in this thesis to optimize its capbility. First, the low-frequency data are adjusted to the range of [0,255] and then saved directly. Second, a marking martrix is set to track the coding position, instead of the dominant table and subsidiary table. Meanwhile, the dominant and subsidiary scans are designed to act on each important coefficient simultaneously. Furthermore, one more bit of precision is added to each important coefficient in the process of subsidiary scan to increase the effective data in each scan. 3) In the hybrid MC(motion compensation)/DCT and wavelet combined video coding algorithm, we use only 4×4 block matching to speed up the searching process. In order to avoid the problem of falling into local minimum that exists in traditional fast searching algorithm, we propose several solving motheds. First, determine whether the block is a static block. If it is static, stop searching of this block. If not, do motion vector prediction to determine a searching center, from where the search begins and the route will be from the inside out, meanwhile, a stop condition is combined to stop searching in an early time and an acceptable position. The I-frames are coded by embedded zero tree wavelet, for the P-frame, an inter video prediction is done and the residue data are transformed by a 4×4 integer matrix, finally Columbus code table is applied to output the bit stream.In the original remote vision control system, the robot arm positioning uses grayscale template matching. This method is simple, but unstable against illumination and background change and shadows shifting. The mismatch would lead to large errors and which made the controller overmodulating and could not draw curves according to the predetermined trajectories. In this thesis, we consider the feature that the gradient derection angles are stable against changes of light and shadows, and apply a 2-D gradient derection coding matching methods to do robot arm positioning. To the problem of large quantitiy of global search, we introduce Kalman filter to predict the location of the robot arm. As the predicted location and actual location are very close, the searching could done in a smaller range, which greatly reduces the computational complexity. The proposed positioning method is applicable to general grayscale video sequences when there is no rotation and the objects are regid.