| Summary: | 碩士 === 長庚大學 === 機械工程研究所 === 92 === Vehicle monitoring and identification has been an important issue in social security and many other applications, such as parking area administration and electric tollgate in free way. Most of the researches focus on name plate identification so far. These approaches suffer from the possibility of replaced name plates on illegal vehicles. We alleviate this problem by multiple-feature identification in this thesis.
This work aims at automatically acquire and recognize multiple features of vehicles. These features include name plate, body color, and brand mark. At the beginning of each of the feature identification processes, specific regions are located by Region Boundary Detection. In name plate recognition, the region is further divided into digit sub-regions. The digits are recognized using a hybrid methodology of the Principle Component Analysis (PCA) and artificial neural networks. In body color identification, the original RGB image representation is transformed into HSV to reduce the lighting effect. In brand mark recognition, the boundaries of the pin-downed regions are traced to locate vehicle mark, and the PCA is used.
According to experimental results of 220 practical images, the system achieves a recognition rate of 95%.
|
|---|
