Design And Application Of Moving Object Recognition And Three-dimensional Positioning Model Based On Monocular Vision

In the context of the rapid development of various types of artificial intelligence technologies,the intelligent upgrade of enterprises is urgent.Among them,the mature target detection and positioning technology enables decision-makers or intelligent analysis systems to accurately analyze the current situation and take next steps through the collected information on the environment,personnel,vehicles,and equipment,which is the basis for future intelligent enterprise production management.However,most of the current target supervision systems have problems such as low detection accuracy,limited application scenarios,high cost of positioning equipment,and the lack of close integration of detection and positioning,and the lack of a complete process model,which hinders the improvement of production efficiency.Aiming at the above problems,the main research work is as follows:(1)The target detection method based on deep learning is studied.According to the image features of people and vehicles in multiple scenes,the YOLO-V3 network model is optimized by adding a spatial pyramid pooling module and introducing a 4-fold downsampling scale,which improves the detection accuracy.Finally,the model is trained and tested using the home-made target dataset.(2)According to the different ranging principles and application scenarios,four monocular vision ranging algorithms are designed based on the actual width of the target,the target detection area,the imaginary horizon and the target ground projection point.Through comparative analysis experiments,the ranging algorithm based on the target ground projection point is selected for positioning research,and the three-dimensional coordinates of the target in the set coordinate system are obtained.(3)A low-cost target recognition and three-dimensional positioning model that meets the current application requirements is proposed,and the model application scheme is designed from three aspects:multi-target precise positioning,single-target path monitoring and target violation and dislocation alarm,and application experiments are carried out.The experimental results show that:under the high IoU threshold,compared with the original model,the optimized YOLO-v3 model has significantly improved target recognition and detection accuracy,up to 83.3%.At the same time,the model runs at a frame rate of 28 frames per second,the real-time performance is better.In addition,the algorithm’s ranging error for vehicles and people in the detection area is basically lower than 4%,which meets the requirements of three-dimensional positioning.In the end,the designed application scheme achieves the expected application goal,which has certain practical significance for promoting the process of intelligent production and management of enterprises.