Scene adaptive target positioning method and system based on spatial perception

Abstract

The invention relates to a scene adaptive target positioning method and system based on spatial perception. The method comprises the steps of: acquiring a target center point quadratic curve and a Gaussian distribution curve of the detection frame size according to a previous target positioning result of a current scene; acquiring a target image in real time, acquiring a plurality of assumed target center point combinations from the target center point quadratic curve according to a target distribution rule, further acquiring corresponding assumed detection frame combinations from the Gaussian distribution curve of the detection frame size, and calculating scores of the assumed detection frame combinations through using an energy function; selecting the assumed target center point combination with the highest score and the corresponding assumed detection frame combination as an auxiliary positioning result of the target; and feeding back the auxiliary positioning result to a deep learning model, and performing parameter adjustment so as to perform target positioning at the next moment. Compared with the prior art, the scene adaptive target positioning method can enable the model to carry out self-adaptive learning according to different fan environments on the basis of a universal deep learning model, and further improves the compatibility and robustness.

Description

technical field [0001] The present invention relates to the technical field of target positioning, in particular to a space perception-based scene adaptive target positioning method and system. Background technique [0002] After years of rapid development of wind power technology, the market has become very large. According to statistics, there are currently more than 70,000 wind turbines in operation for power generation, that is, more than 200,000 blades are in operation for power generation. Among these blades, about 0.05% of the blades will be broken and damaged due to various reasons every year. All blades must go through various experiments and obtain relevant qualification certification before mass production. Therefore, as long as there are no major quality defects, there will be no fracture accidents in the early stage of the blade's on-hook operation. However, once the blade operation enters the middle stage, under repeated fatigue loads, the remaining manufact...

AI Technical Summary

Problems solved by technology

However, the image data obtained in different wind turbine environments are also very different. Since it is difficult to obtain a large amount of data from each wind turbine to train the model, the generality and robustness of the model are very limited.

Images