AM-GPD: Manipulator Grasping Pose Detector Based On Attention Mechanism In Dark Light Scene

To implement a competitive grasping pose detector in dark light scenes, an attention-based mechanism is proposed. The basic idea can be described as follows. First, the attention module is used to accurately identify the dark regions of the region of interest. By using an image brightness enhancement method, the brightness of the dark region is increased to the desired level. Subsequently, an efficient residual convolutional neural network model based on the attention mechanism is proposed to estimate the grasping pose of the target object in the visual field. An optimal object grasping pose is derived by using 640x480 RGB maps and depth maps as input to generate mass, angle and width maps. Simulations of the Cornell and Jacquard crawl datasets were performed in the dark light scene. The experimental results show that the detection accuracy of the proposed grasping pose model in this paper reaches 98 % and 92 %, respectively.