The Real Time Object Detection System Based On BP Network

Human being is always the key role in watching certain palaces especially those visual monitor aided by cameras. The qualities of monitoring these places vary a lot in some cases, for instance, the shifting of surveillant or the same person in different time. This variation could be eliminated if we introduce computer to this kind of tasks, but at the same time it also have to face some newly introduced troubles brought by computer vision. For example, the popular way of detection is background difference method, but this method has its own flaws that it can not reflect the real-time changes. For instance, the variation of illumination intensity, the movement of our uninterested background object or the rain drops on the water surface could not be interpreted into a single background. If we could not handle this, we could not separate our interested object from the video sequence, and this would significantly degrade the recognition percentage which makes our detection useless for further use.In order to remove this kind of trouble, a new method called self-adaptation method has been introduced into detection. But it still needs to be improved a lot. In my dissertation, I use Back-Propagation neural network to remove the instability and complexity of traditional self-adaptation method. My work is primarily concentrated on following places:1. Solve the problem caused by the massive input of BP network.2. Decide the structure of BP network through experiments.3. Test some popular ways of training method and choose Variable Learning Rate Method as my training method.The experiment data shows that it is possible to use BP as the core algorithm of self-adaptation method, and the most important thing is that it is easy to adjust to a new place without any expert.