Unlike human vision,which can adapt to various lighting environments,the performance of existing 2D human object detection algorithms will be significantly reduced in the scene of drastic lighting changes.In order to solve this problem,in this paper a two-dimensional human object detection method was proposed,which adapted to the environment of drastic lighting changes.Firstly,based on the stage performance environment with drastic lighting changes,a human body image benchmark dataset(named"StageHuman")containing various light colors and rich light changes was collected and constructed to verify the defects and deficiencies of the current two-dimensional human object detection algorithm.Secondly,a data enhancement strategy based on style transfer was proposed,which migrated the dramatic lighting change style in a specific scene image to the body picture of a large-scale open data set COCO,and then used the large-scale data set after style transfer to train the deep neural network model,so as to improve the two-dimensional human detection performance of the model under the environment of dramatic lighting change.Finally,through a large number of experimental comparison and analysis,it is verified that the proposed method can effectively improve the robustness and detection accuracy of the deep neural network model under the environment of drastic lighting changes,and the effectiveness does not depend on the specific style transfer algorithm,but mainly depends on the diversity and integrity of the lighting change styles transferred.
维普
万方数据