Box Detection and Positioning based on Mask R-CNN [1] for Container Unloading
Zhangli Zhou, Meiling Wang, Xiangyang Chen, Wei Fang Liang, Jun Zhang · 2019
Container unloading is a very dangerous job, so how to automate it is very necessary.There are currently machines that can be fully automated and unloaded for specific situations, but the generalization is not good. In order to automatically unload the goods under different situation, it is necessary to find a way to replace the human eye, how to establish a complete identification and positioning Vision system is very important.Traditional methods based on Vision are difficult to achieve end-to-end unloading automation.We use deep learning-based target detection here. The algorithm Mask-RCNN is used to detect specific objects. The second generation Kinect developed by Microsoft can completely re- place the human eye. In the simulation environment, we get the recognition speed of each frame for 0.2 seconds, which can get the three-dimensional coordinates of each box precise. By feeding back the data to the machine, the machine can know their target clearly.This is a good promotion of the development of automatic unloading.