行車紀錄器應用於車款辨識系統

在本篇論文中，我們設計了一個車款辨識系統，應用於行車紀錄器上進行辨識車款資訊。我們設計此系統目的是為了可以讓行車紀錄器上可以更精準的紀錄更多詳細的資訊，而不再只是單純影像紀錄。本研究分為四個部分，第一個部分我們將行車紀錄器中的影片，擷取成影像做後續的影像處理。第二部分將擷取下來的彩色影像從左至右，從上到下分割成9宮格，再將位於中間格子的影像進行HSV處理，選取符合預設的長寬比與面積的區塊進行篩選，並針對所需要的區塊進行編號。第三部份依據已編號的紅色圖像進行(Region Of Interest, ROI)，也就是車子的尾燈部分進行特徵點的擷取計算。最後一部分則利用，KNN(K Nearest Neighbor)演算法的方式，達到整體辨識的實驗結果。本篇論文的研究結果貢獻如下： 1.降低開發成本由於本文是採用中階款的行車紀錄器進行模擬研究，無需使用到工業用等級的高階攝影機進行擷取，因此在各種硬體技術上的接受度相當高。 2.提升準確性經由判讀後可以於行車紀錄器上記錄前車更多的資訊，讓記錄本身更加完整。

關鍵字

車款辨識

並列摘要

In this thesis, we will design a car identification system that can be used on the driving recorders to identify vehicle information. The purpose of our design is to make it more accurate to record more detailed information on the driving recorders than just a simple image recording. This study is divided into four steps. The first step, we will intercept event data recorder films into images to make the subsequent image processing steps. The second step is catch images from sequence, and split link image to nine parts, from left-up to right-down, and catch middle-center part to HSV conversion, then select the block which is conform the set that area is reasonable and ratio of height and width, and give these area an number. The third step is to accord to that numbered red parts, and calculate the feature points on car taillight. At last step, we use feature classification to select the car type, and explanation the results of overall identification. The research results of this thesis contribute as follows: 1.Reducing development costs: Because this thesis using the mid-range driving recorders for simulation research, it does not need to use industrial-grade high-end cameras for capturing, so that the acceptance of various hardware technologies is quite high. 2.Improving accuracy: After the interpretation, you can record more information on the preceding vehicle on the driving recorders to make the record itself more complete.