| As the Chinese economy develops,the rail technology advances and its scale has expanded,boosting the efficiency of the transportation of both supplies and personnel.To meet the increasing demand,safety emerges as a great concern as the intrusions,including the objects in the rail and the pedestrians beside the rail,pose potential challenges to the smooth operation of the rail system.However,the traditional detection method is highly resource-consuming and unable to be fully-automated.In comparison,the video surveillance,introducing more possibilities into the detection of the foreign object intrusion,offers wider coverage and lower costs and is more suitable to monitoring in an open environment.Yet many factors could contribute to the low resolution and data lose,exerting negative impact on the downstream practices.The present research studies the detection mistakes and omissions induced by low resolution and the major content of the thesis is as follows:(1)Propose research questions and goes on to review the research background of rail foreign object intrusion and Super-Resolution reconstruction algorithm.(2)The study has reviewed the most employed methods,their processes as well as their advantages and disadvantages so as to make improvements.(3)To enhance the resolution,the study had modified the generator,discriminator and loss function of the original network in order to raise the image evaluation indexes,including Peak signal-to-noise Ratio and Structural Similarity.(4)Employs SSD(Single Shot Multi Box Detector)object detection algorithm to test the design.According to the results,while the detection omissions occurred under low resolution,more smaller objects could be detected under high resolution.Considering that the smaller objects only appear in the outside-platform area,the study modified the detection system with OPENPOSE.The method differs from SSD since it can conclude whether the foreign object is a person,instead of offering a percentage of probability. |
PDF Full Text Request