DeeplabV3 +-based instant positioning and mapping method for obstacles in unfamiliar sea area

Abstract

The invention discloses a DeeplabV3 +-based instant positioning and mapping method for obstacles in an unfamiliar sea area, and the method comprises the steps: firstly, projecting a laser point cloud to an image based on a conversion relation between a laser radar coordinate system and a visible light camera coordinate system, and obtaining a laser point cloud projection coordinate in an image coordinate system; performing segmentation prediction on the image based on a Deeplabv3 + semantic segmentation network model to obtain a segmented category image, making an obstacle mask by using an obstacle category to obtain a laser point cloud corresponding to an obstacle, and finally, according to a conversion relation between a laser radar coordinate system and an IMU coordinate system, obtaining a laser point cloud corresponding to the obstacle. And constructing a surrounding obstacle map taking the unmanned ship as the center of the map according to the coordinates of the laser point cloud corresponding to the obstacle in the laser radar coordinate system and the attitude information of the unmanned ship acquired by the IMU. According to the invention, accurate extraction of predictable and unpredictable obstacle contour information is realized by using a Deeplabv3 + semantic segmentation network model, and accurate detection and positioning of offshore predictable and unpredictable obstacles are realized in combination with a visible light camera and a laser radar.

Description

technical field [0001] The invention belongs to the technical field of unmanned boats, and in particular relates to a DeeplabV3+-based real-time obstacle positioning and map construction method in unfamiliar sea areas. Background technique [0002] Unmanned boats have been used more and more widely in recent years. For example, in the military field, unmanned boats can be widely used in ship protection, anti-submarine warfare, logistics supplies, information investigation and other fields. In the civilian field, unmanned boats can be widely used in Marine environmental monitoring, seabed surveying and mapping, emergency response and sea area search and rescue work and other fields. Since the unmanned boat and its equipment are often expensive, if it collides with objects in the ocean during navigation, it will seriously damage the unmanned boat and equipment. If it collides with a ship, the consequences of injuring people may be even more unimaginable. At present, technical...

AI Technical Summary

Problems solved by technology

At present, the detection of marine obstacles generally adopts the YOLO series network model, which has good detection results for predictable obstacles, but cannot identify unpredictable obstacles; the positioning method of marine obstacles mostly uses laser radar or Binocular camera positioning method

The laser clustering method is easily affected by noise, has a high false alarm rate, and for large obstacles, using the clustering method takes a lot of time

The binocular positioning method has poor stability due to the long baseline, and poor matching accuracy for reflective objects and weak texture obstacles, and the positioning accuracy is not high

Images