PYNQ-based Map contours’ recognition and mosaic device
Guo Yu,Yao Weizhuo,Shen Zhengguo
In traditional image recognition, image Mosaic is generally based on image content, color and other elements, and the matching degree is calculated by algorithm. However, this method is obviously unable to deal with a situation: when there is no feature graph for recognition inside the image, content-based recognition cannot accurately locate and calculate the order and arrangement of image fragments. Aimed at this situation, we used the method of image contour feature extracting, all data based on the outline of the image, is no longer dependent on connected domain image color information, on the one hand, to solve the deficiency of the original algorithm in the case, makes the result more accurate and reliable, on the other hand, due to save the data only image contour information, reduce the pressure of the hardware computing, make the system more smoothly fast. In the future, it can be used in recognition and sorting of factory parts, moving detection of specific targets and other aspects.
The system consists of pynq-z2, USB camera, hdmi display, motor drive, transistor diffuser and power management module. The power management module is mainly used to supply power to each submodule. First obtain image‘s information by the camera, and then transfer the information to the pynq’s ARM kernel to process it, after that, control instruction issued by the FPGA to control motor and servo’s movement and rotation and electromagnet’s switch, to send corresponding puzzle pieces to the specified location, at the same time the results of image processing will be output to the screen by the hdmi output of the board.
Major innovations points of our work:
1. Different from the general algorithm which matches the puzzle by color proportion, we extract the outer contour of the puzzle (the inner contour is filtered out) to calculate the matching degree of the contour, so that we can achieve a very high accuracy matching when facing the image with complex edges.
2. A feature triangle is formed by determining the center point of the image and the longest feature line, so as to determine image’s absolute Angle of rotation. By contrast, the general algorithm to calculate the Angle by taking the minimum external rectangle can only calculate the horizontal Angle of the edge closest to the horizontal axis, which does not meet our requirements.
3. The hardware structure is designed and spliced by ourselves, including x, y, z and rotation dimension w to realize the restoration and splicing of the puzzles.