| Summary: | 碩士 === 臺灣大學 === 電子工程學研究所 === 97 === In this thesis, we propose a method based on image processing to develop a low cost night vision system for vehicles. The system captures the road situation at night with visible light video camera and provides the driver with the enhanced clear video. In our method, our proposed method include two parts, denoising and enhancement. In denoising, we segment the image into dark part and bright part by average brightness. Calculate the thresholds and use sobel edge detection to do edge detection, respectively. After getting the edge map, we perform edge-preserving spatio-temporal filter. In enhancement, we modify local gamma correction. Replace the exponential function in it by linear function with dynamic slope. Divide the image into dark part and bright part by the value we set. Brighten the dark part and darken the bright part. Moreover, we use the concept of retinex to modify local contrast.
Finally, We implement our algorithm in FPGA chip, the clock frequency is 54MHz. Inside the chip, we use line buffer to store the temporary date. We design the control circuit to let the date get in and out regularly. This will make the hardware easy to realize.
|
|---|
