Study On Automatic Door Anti-pinch System Based On Video Analysis

Automatic doors are used more and more widely today with the rapid development of the elevator, subway, etc. How to achieve automatic door's anti-pinch feature to ensure security of the process of running automatic door, ensure personal and property safety has become a research hotspot in the field. The traditional anti-pinch system of automatic door is mainly based on the touch pad or infrared screens, however, pre-judgments can not be achieved in the former type, resulting in frequent accidents, less secure and the infrared transmission lines of the latter type are vulnerable to interference, resulting in error-prone judgments.Machine vision has become more sophisticated in recent years, and application system based on machine vision has advantages such as high speed, high precision, non-contact and so on. In this paper, the system takes full advantage of these features to analyze whether there are obstacles appearing at the door according to the video, properly control automatic door's movement. Finally, the system achieves the purpose of anti-pinch, so as to effectively overcome the shortcomings of traditional anti-pinch methods.This thesis mainly studies obstacle detection technology based on analysis of automatic door video, and how to build the anti-pinch system on DSP platform according to image processing technology. First, study image preprocessing method combining the image characteristics. Propose the algorithm of adaptive median filter which can remove the salt and pepper noise and so on effectively, based on the statistical properties of local image pixels. To meet the needs of analysis, using normalized cross-correlation algorithm to locate ROI area accurately. Second, do research on image segmentation. The foreground through automatic doors constantly changing determines a higher demand on the segmentation threshold, so methods on segmentation criteria and zoning on the two-dimensional Otsu segmentation algorithm are improved, which can improve segmentation accuracy while reducing the amount of two-dimensional Otsu calculations significantly and separate out the feature regions of automatic door bottom accurately. Finally, using sequential algorithm to label target regions in binary image, while calculating image features like centroid and long axis. According to the particular light environment of automatic door, the shadow processing algorithm based on the vertical gradient of edge is proposed, which can distinguish shadow and obstacle effectively in images to reduce the anti-pinch system malfunction.This system is tested at algorithm level and program level using Simulink platform and DM6437 DSP platform. The video-based anti-pinch system of automatic door is not only significant in theory but also strong valuable in use. On this basis, further development of more rapid and accurate detection and the economical hardware platform has a very broad prospect.