Automatic Container Recognition and Positioning Method Based on Hough Transform and Mask RCNN

Abstract

In multimodal transport, accurate identification and positioning of container is the key to construct container yard map. However, container recognition accuracy is low and vulnerable to the environment using the traditional Hough Transform. This paper proposes a container automatic recognition and positioning method based on Hough Transform and Mask Region-based Convolutional Neural Network (Mask RCNN) algorithm. The method consists of two parts, pre-processing of container images and instance segmentation using Mask RCNN algorithm. In the pre-processing part, Hough Transform is used to detect the polygon contour lines of the container image, and then the contour lines are filtered according to the container contour features to locate the container initially. In the segmentation part, using Mask RCNN algorithm, container features are detected for the pixels within the target contour line to identify the upper surface contour of the container, thus the exact location of the container is determined. The experimental results show that the method improves the recognition effect of traditional image processing algorithms and increases the stability and accuracy of container recognition and positioning.