The Research And Implementation Of Image Procesing Based On OMAP-L138

With the development of society and technology, the demands of image information are growing fast in people’s daily life, which makes the image capture technology develop rapidly. At the same time in intelligent transportation systems in order to solve problems such as the traffic jams, road parking, traffic accidents the license plate recognition technology has also got rapid development. The license plate recognition technology is widely used in highway intersection charge system and public parking charge system. On the basis of obtaining license, license plate recognition technology is composed of license plate image preprocessing, license plate location and license plate segmentation.OMAP-L138 EVM is used as an embedded system development platform in this thesis. The OMAP development platform is widely used in the multimedia application due to its high performance, low power consumption and high integration. OMAP-L138 is based on ARM and DSP dual-core processor architecture, and the seamless connection between dual-core can be realized while using the Codec Engine software framework provided by TI implements. ARM is mainly responsible for the upper application and the task scheduling, and DSP side is responsible for tasks of real-time processing or large computation. As DSP and ARM system are controlled independently under dual-core processor architecture, code execution efficiency is highly improved and power consumption is greatly reduced.In this platform USB interface is used to connect the UVC camera as the image capture device and the image is captured under the V4L2 software framework. At the same time, the captured image is compressed by JPEG encoding. On the basis of obtaining the license plates RGB data through the JPEG decoding, license plate can be located by image gray scale, image binarization, image denoising and image edge detection algorithm. The character is split by vertical projection method, and then normalized by method of gravity normalization. At last license plate recognition is completed after analyzing the difference of character recognition algorithm based on template matching and BP neural network matching algorithm. The results shows that the BP neural network matching algorithm is more accurate and has better performance on time than template matching method. Under the guidance of the rules and guidelines provided by TI the license plate recognition algorithm is encapsulated as library function, and can be called through Codec Engine framework, and then finally the license plate recognition algorithm can run on DSP side. This makes a further optimize on the time performance of algorithm, and also reduces the difficulty of transplantation under other platform.