Research And Implementation On Real-time Foreground Target Detection Based On DM6437

As a critical step in video image processing, moving target detection has become an important topic in the field of computer vision. With the widely application of intelligent monitoring system, embedded video processing system which with a portable and low cost advantages also get rapid development in recent years. As embedded processing platform such as DSP always with less memory resources, the process are more single, so it is very important to develop high accuracy and real-time detection algorithm.To improve the operating efficiency of foreground detection algorithm, this thesis proposes a method which mix the super-pixel and CodeBook background model to detect moving target. The algorithm shows good detection effect when we analysis the result of simulation. In addition, this thesis designed and implemented an embedded processing system based on DM6437. The algorithm works real-time in the embedded platform and also has good performance. The main research contents are as follows:1. To improve the disadvantages of CodeBook algorithm in detection speed and memory consumption, this thesis proposed a method mix super-pixel and classic CodeBook detection algorithm. The background modeling of this algorithm based on area which has same characteristics such as color and texture. Then target detection is completed according to the regional characteristics. The result of target detection of this algorithm preforms slightly better in three standard data sets than classic CodeBook algorithm and grid CodeBook algorithm which are used commonly. Besides, the efficiency of detection is superior to other algorithms.2. A hardware platform based on DM6437 is designed and implemented. The platform is designed by eight layers. DM6437 is chosen as the main processor. The platform provides a rich memory space and external interface, also has a faster data transfer rates. The hardware platform can runs stable and process video data real-time.3. According to the characteristics of the DM6437 hardware platform, the thesis designed a transplantation and optimization strategy of foreground detection algorithm based on CCS integrated development environment. This takes up less system memory and operates more efficient and stable.4. This thesis completes target detection on different scenes which include indoor and outdoor, light and shade based on the embedded system platform which is based on DM6437. The result shows that this target detection algorithm we proposed which mix super-pixel and CodeBook preforms better in the detection effect, the detection efficiency and memory consumption than the classic CodeBook algorithm.So this thesis illustrates that the algorithm has a lot of advantages in the application of the embedded video processing platform.