Application Of Binocular Stereo Vision In Elevator Safety Detection

Elevator is an indispensable tool in people’s lives,with the increasing of usage of elevator,elevator safety accidents are more frequent too.So the safety detection of the elevator is a contemporary urgent demand.In the elevator safety detection,the state of the elevator door is a key detection project,and binocular stereo vision technology with its non-contact,passive,high accuracy and real-time characteristics,is very suitable for the range/speed measurement of elevator door opening and closing.In this paper,stereo vision technology and image processing technology are applied to complete the range/speed measurement of elevator door opening and closing.The main work is as follows:(1)The image and video capture of binocular stereo camera,binocular stereo calibration and binocular stereo rectification are completed.The binocular stereo vision measurement model and its geometric transformation are analyzed.The theory of feature extraction and stereo matching is analyzed.And the binocular stereo vision ranging function is completed.(2)Using the frame difference method,the rough position recognition of the elevator door is preliminarily completed.And then put forward three methods to improve the recognition accuracy of the elevator door edge:The first is the Otsu threshold segmentation method;The second is vertical edge recognition method based on the horizontal gradient,which weakens the interference of the other direction edge;The third is a combination of the above two methods.(3)The motion prediction model based on polynomial fittingis applied to the elevator door edge detection.A kind of Gauss weighting function is proposed to describe the confidence of prediction.Using this Gauss weighted function,the paper combines the edge detection method in the previous one with the motion prediction model,then a hybrid algorithm of edge detection is proposed which combines Otsu threshold segmentation and horizontal gradient edge detection first,then weighted with Gauss function at the predicted location.(4)The problem of stereo matching and 3D reconstruction is studied.Then two kinds of general stereo matching methods are proposed,which are real time region matching and dual video recognition simultaneously.And a single time matching method in the special case of the camera plane parallel to the elevator plane.According to the characteristics of the opening and closing of the elevator,the paper puts forward a variety of constraints which are helpful to improve the performance of the algorithm.