Implementation of a bacterium tracking system on FPGA

Tracking in computer vision is monitoring the movement of an object in a sequence of images. This object tracking may be performed for different reasons such as security systems, animation production, etc. Hundreds of publications in the form of books, articles, and journal papers indicate the importance of tracking as the research domain in universities and research centers. Tracking consists of the answer to two principal problems, the motion problem and the matching problem. Two solutions for the motion problem are adjacent regions method and the Kalman filter. Many solutions also exist for matching problem such as window tracking, detection of the moving object by its specifications like edges, corners and contour, detection of the target by simple shapes like circles, squares, rectangles and triangles, detection by 3-dimensional shapes like cylinders and finally matching with complex shapes like human, vehicle, etc. The implemented tracking algorithm in this project applies the adjacent regions for motion detection and window tracking for matching. Three different algorithms are simulated by Matlab software and one of them is implemented on the target hardware.;The subject of tracking in this project is Magnetotactic Bacterium (MTB). MTB is applied in medial and biomedical fields. As an example, they can carry necessary materials into the patient's vascular system, where only nanoparticles are able to travel.;The hardware of this project is the Xilinx ML402 video starter kit, camera and microscope. This hardware system consists of the M L402 main board and the VIODC daughter board. The VIODC contains a Virtex-II FPGA which holds the drivers of different video input and output interfaces. The M L402 has a Virtex-4 FPGA and developers can implement their designs on this FPGA chip. The camera captures the video frames and the VIODC receives the frames from camera and sends them to M L402. ML402 executes the tracking algorithm on the received frames and sends the appropriate output to VIODC to be displayed. The basic module of the tracking system which is called "vid_tracking" processor core (pcore) is available with the Simulink library of the Matlab software. It can also be simulated either by real-time video or by simulated video signals. Vid_tracking consists of EDK processor and System Generator blocks. System Generator translates the pcore to physical hardware. Through the translation process, the EDK processor of Simulink library is implemented as a MicroBlaze processor which is the standard processor architecture of Virtex-4 FPGAs. The design is then exported into the XPS platform. The system developer can modify the design using the VHDL hardware description language. VHDL is very efficient and highly flexible to implement hardware systems with different abstraction levels.;We applied three DSP48 modules from the Virtex-4 library in order to implement the multiplication operation in a more efficient manner. This library contains some already designed components for different Xilinx FPGAs. These components are very fast and easy to use. In order to implement the tracking algorithm, we profit from three synchronization signals which are pixel enable, horizontal synchronization and vertical synchronization. Pixel enable indicates when a new pixel is detected by the VIODC. The horizontal synchronization signal indicates the start of a line and the vertical synchronization shows the start of a frame. The tracking algorithm counts the pixels and lines to find a window of about 20-pixel by 20-pixel in the middle of the frame. It looks for a bacterium by comparing the pixel's intensity. When it finds a bacterium, it looks for it in the 20-pixel by 20- pixel regions around the previous window in the next frame.;The Matlab simulation of MTB tracking is 100% robust since the bacterium was not missed in any of the frames during the tracking and the algorithm is able to follow the trajectory of the bacterium during all the simulation time. The simulation indicates 100% precision in 2- dimensional movements of the bacterium, since all the pixels which are recognized by the human eye as the bacterium pixels are also detected as the bacterium pixels by the simulator algorithm.;The designed hardware detects whether a bacterium is in camera field of view or not. The bacterium moves at an average speed between 180 to 240 micrometers per second. A speed of 300 microm/s has also been recorded for this type of bacterium. Our tracking system could detect the moving bacterium at either of these speeds.