The Automatic Detection Of The State Of Locomotive Roof

The detection of the state of locomotive roof is one of the most important parts of locomotivedetection. With the speed of locomotive raised and burden increased, one of the major concerns is thesecurity of the locomotive. Locomotive roof cannot be checked easily, however, there will be a lot ofinsecurity factors on the roof after the locomotive has been running for a period of time, such asforeign objects, damaged key components such as porcelain and pantograph, which seriously affectthe running safety of the locomotive. At present, there is no complete set of equipment to check theroof automatically at home or abroad. Hence, in this thesis, the state of locomotive roof has beenstudied in detail.In this thesis, the automatic detection of the state of locomotive roof is divided into the followingparts: foreign object detection, localization and detection of key components, cleanliness stateanalysis. Firstly, it needs to generate panorama of locomotive roof without distortion. Since the largescene images taken by area array cameras are severely distorted and not easy to be recovered, severalline scan cameras are used to acquire the images of locomotive roof. A speed based roof imagecorrection method is proposed to deal with the longitudinal distortion caused by the changing of thespeed. And then, the corrected images can be mosaicked to a panorama of locomotive roof quickly,based on the pyramid structure multi-resolution operator. The result shows that the large scene imagemosaicked using this method has the advantage of low distortion and high resolution, which increasesthe precision of detection of foreign objects and key components. For foreign object detectionproblems, SIFT feature vectors of the image is extracted firstly. To speed up the image processingspeed, image will be divided based on feature points. And then, improved RANSAC operator will beused for template matching. Finally, the detection will be completed based on the improvedsymmetric differential algorithm. The improved algorithm has the advantage of low mistake ratecompared to ordinary differential algorithm. For key components localization and detection problems,all components are located accurately in this thesis, based on template matching. And then thecomponents are detected via edge detection algorithm and Hu discrete invariant matrix. The resultshows that this detection method is less affected by light and has a high detection precision. In theanalysis of the state of cleanliness, improved Kirsch fast operator is used, which has smaller amountof computation compared with conventional Kirsch operator.In this thesis, the proposed automatic detection method of the state of locomotive roof is realized in simulation experiments, and it can meet the real-time and accuracy requirement in the detectionsystem. The technology has some application prospect in the detection segment of low speedlocomotives.