Research On Facade Feature Recognition And Job Effect Detection Based On Binocular Vision

With the rapid development of wall-climbing robots,wall-climbing robots have also been introduced to complete tasks for large facades such as ships and tanks.At present,manual operation of wall-climbing robots has many disadvantages such as low efficiency and inflexibility,and high hazard coefficients for aerial work.How to achieve flexible and efficient facade operations has become an urgent problem to be solved.In response to this problem,the automatic facade operation and effect detection of the wall-climbing robot will be the key to solving the problem.Wall-climbing robots are unable to automate facade operations mainly due to the presence of facade features on the surface of ships and storage tanks,such as flanges and bolts.Because the wall-climbing robot cannot identify and locate its specific size,it will not be able to work on it,and can only avoid it.With the rapid development of computer vision,all fields have been well applied.Therefore,in this paper,the binocular vision system and structured light system are used to identify the fa?ade features of ships and storage tanks and the effect detection after operation,and they are carried on mobile robot platforms for application.The specific research contents are as follows:First of all,according to the facade feature database of actual ships and storage tanks,the visual design is completed,the simulation facade feature database is constructed,and the software and hardware platform construction and image preprocessing of binocular vision are completed.Secondly,the research on the facade feature recognition algorithm based on binocular vision was carried out.Zhang’s calibration is used to obtain the camera parameters,and the binocular images are subjected to stereo correction based on the Bouguet algorithm to construct the stereo matching parameters.The SGBM stereo matching algorithm is used to complete the outdoor environment.The parallax of the facade features is restored,and the distance measurement is completed using the principle of triangulation.Then,based on the SSD target detection algorithm,the facade feature detectio n in the outdoor simulation environment is realized,the facade feature images under different weather and lighting environments are collected,the facade feature data set is created,and the facade feature recognition model is trained.In view of the redundancy of the classic SSD model,the convolution kernel pruning method is used to optimize the model.When the recognition accuracy of the model is not significantly reduced,the model speed performance is improved to achieve the expected effect.In order to improve the automation of facade operations,a binocular structured light vision platform is built,common coding methods is analyzed,spatial coding methods are used to complete the depth measurement of facade features,and analysis of accuracy and influencing factors.Finally,the visual mobile platform is built,the visual system is placed on the mobile platform,and the visual program is transplanted to NVIDIA TX2 embedded development board.The experiments of static ranging,real-time dynamic recognition and facade operation effect detection are carried out.The results show that the binocular vision facade feature algorithm and facade operat ion effect detection algorithm adopted in this paper can efficiently and accurately identify and detect the facade features in the outdoor environment,and can provide technical support for the automatic facade operation of wall-climbing robots.