Image Processing And Its Implementation On Hardware For An In Situ Micro Particle Observation System In Deep Sea

71% of the earth surface is covered by sea water, the response from the ocean due to global climate change has triggered a direct impact on our society, economy and daily life. Hence, there are many associated scientific problems worth exploration. Nonetheless, we are short of knowledge of the ocean, especially the deep ocean. This shortage becomes a driving factor for development of advanced instruments. These instruments can be deployed in deep sea for in situ observation, a better theory of the nature law can be established based on the data collected by them. In this context, it is reasonable to design an in situ particle observation system in deep sea. This system can observe the movement of sediments and/or algae in seafloor and help to improve our understanding of the rule behind the flow and material exchange in deep sea. Such instrument offers a new possible angle that contribute to our progressing knowledge of the ocean.The system is based on high-speed high-resolution image capturing, typically it can work in 2000 m deep sea, achieve 500 fps at a resolution of 1280*1024. This dissertation studies the core algorithms and their implementation on hardware for an in situ micro particle observation system. A wide range of work is covered, from theoretical analysis to running codes, experiment validation, and implementation on hardware. The image processing mainly consists of two parts: particle detection and velocity calculation. The principle of these two algorithm is analyzed and customized, then experiments in a sink are carried out to validate it. This dissertation also focuses on the algorithm migration from simulation on a PC to implementation on a smart image processing chip. The major results achieved by this dissertation includes:1. The process to detect particle is established based on integrating and improving current algorithms. Data transmission is sharply reduced by effectively extracting meaningful information. 625MB/s of the original image data are transformed to less than 4MB/s of particle information, a reduction of 2 orders. The new algorithm is capable to handle the huge amount of data produced instantaneously by high-speed high-resolution camera, and to detect particles in real time.2. The algorithm to calculate particle velocity is implemented. Based on traditional Particle Tracking Velocimetry, the new algorithm is customized for high-speed high-resolution camera. The probability matching situation is simpilified, only 2 cases are considered, and iteration time for the complicated one is reduced. Velocity can be calculated in real time and velocity field can be presented.3. These two algorithms are validated by experiments on a sink. A special sink is designed and the velocity distribution is calculated theoretically. By adding given particles into the sink, taking images under different flow condition, parameters are optimized, and further improvement is supported.4. These two algorithms are migrated to smart image processing chips successfully, which narrows the gap from theory to application and makes it possible for practical deployment.