A method for positioning a water gauge of a vessel

Abstract

The present invention proposes a method for positioning a water level gauge of a ship, comprising the following steps: step 1, applying edge detection technology to determine the hull area; step 2, applying a fast regional search algorithm to mark the interest area that meets the color, size, and aspect ratio; step 3, measuring the elliptical properties of the characters in the interest area to preliminarily determine the size, tilt, distortion and other characteristics of the characters; step 4, based on the learning framework of a convolutional neural network and the recognition of printed characters and character libraries; step 5, performing a correlation test on the identified interest area, first arranging the characters in order, referring to the water level gauge similarity matrix weight table, forming a plurality of water level gauge object sets, calculating the degree of familiarity of the water level gauge objects that meet the water level gauge characteristics, and outputting the water level gauge object set with the highest degree of familiarity. This method solves the water level gauge positioning problem under the conditions of severe tilt and severe distortion of the video image of the ship water level gauge and free shooting between 1 and 20 meters of the video shooting distance, and provides a basis for water level gauge measurement.

Description

Technical Field

[0001] The invention relates to the technical field of image recognition, in particular to a method for positioning a water gauge of a ship. Background Art

[0002] At present, bulk cargo delivery is settled by water gauge measurement. Water gauge measurement is based on the average value of six scales evenly distributed around the ship. In this way, the gross weight and tare weight of the ship are measured and calculated respectively, and the difference is the weight of the cargo. Although water gauge measurement is intuitive, it is often affected by factors such as river waves and human visual inspection, which often leads to large errors in measurement results and is easy to be unfair in the settlement process. In addition, the entire measurement process basically relies on manual measurement. On-site measurement and calculation generally take nearly 3 hours, and the low efficiency affects the unloading time.

[0003] Using computer image recognition technology to process video data shot by mobile phones to obtain water gauge readings can reduce the error of manual visual observation. The water gauge measurement process includes four main processes: water gauge positioning, water surface boundary positioning, water gauge reading, and loading and unloading volume calculation. Water gauge refers to the measurement scale printed on the ship for calibrating the draft of the ship. Water gauge positioning is to use video image recognition technology to eliminate various interference factors and identify the ship water gauge image area in the image. This patent mainly realizes the algorithm of water gauge positioning, which is a key step in ship water gauge measurement.

[0004] The water level positioning of image recognition technology is mainly troubled by the following factors:

[0005] 1. Bulk carriers have various shapes and hull colors, and the standards for determining the water gauge of ships are various (continuous lines, discontinuous lines). Other symbols are often printed around the water gauge to interfere with water gauge identification. It is necessary to solve the water gauge position problem of different types of ships;

[0006] 2. Affected by the on-site shooting conditions, the water gauge has problems such as tilt, distortion, and serious imbalance of the upper and lower proportions of the water gauge in the video image;

[0007] 3. Since it is difficult to achieve precision when the ship is berthing at the port, it is impossible to install a fixed camera projection, so a mobile portable terminal is used for shooting. The water level positioning needs to solve the problem of video water level positioning with a spacing of 2-20 meters;

[0008] 4. When measuring with the rear scale (no-load state), the water scale at the bottom of the ship is often rusted and cannot be clearly seen. Summary of the invention

[0009] The purpose of the invention is to provide a safe and reliable method for positioning a water gauge of a ship.

[0010] The technical solution adopted by the present invention to solve the above technical problem is: a method for positioning a water gauge of a ship, comprising the following steps:

[0011] Step 1, applying edge detection technology to determine the hull area;

[0012] Step 2, applying a fast area search algorithm to mark the areas of interest that meet the color, size, and aspect ratio;

[0013] Step 3, measuring the ellipse attributes of the character in the interest area to preliminarily determine the size, tilt, distortion and other characteristics of the character;

[0014] Step 4: Recognition of printed books and character libraries based on the learning framework of convolutional neural networks;

[0015] Step 5, perform correlation detection on the identified interest regions, first arrange the characters in order, refer to the water ruler similarity matrix weight table, form multiple water ruler object sets, calculate the water ruler object's familiarity with the water ruler feature, and output the water ruler object set with the highest familiarity.

[0016] Furthermore, the input of the convolutional neural network is the video image taken by the mobile terminal and the location of the water gauge; the output of the algorithm is the water gauge area image and the water gauge object set.

[0017] Furthermore, the convolutional neural network structure includes three layers, convolution layer one, the first pooling layer one and normalization layer one, the convolution kernel size of convolution layer one is 5×5, the step size is 4, and there are 155 convolution kernels in total; convolution layer two and normalization layer two, the convolution kernel size of convolution layer two is 4×4, the step size is 3, and there are 158 convolution kernels in total; pooling layer three and normalization layer three, the convolution kernel size of convolution layer three is 3×3, the step size is 2, and there are 144 convolution kernels in total; the classifier adopts Softmax classification.

[0018] The advantages of the present invention over the prior art are that water gauge positioning is a key technology for ship water gauge measurement. The algorithm solves the water gauge positioning problem under the conditions of severe tilt and severe distortion of ship water gauge video images and free shooting at a video shooting distance of 1-20 meters, thus providing a basis for water gauge measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0021] like Figure 1As shown, the purpose of the present invention is to provide a safe and reliable method for locating the water gauge of a ship. The technical solution adopted by the present invention to solve the above technical problems is as follows: the algorithm is based on the learning framework of a convolutional neural network and water gauge numerical recognition of a printed book and a character library, and applies edge detection, rapid area search, horizontal and vertical projection interest area segmentation technology, and proposes a character area correlation detection technology to achieve rapid positioning of the water gauge area.

[0022] The positioning process of the water gauge positioning algorithm is: a. edge detection -> b. fast search of character area -> supplementary search of characters in rusted area -> c. tilted ellipse algorithm of character area -> d. character recognition -> e. correlation detection of character area -> f. confirmation of water gauge position;

[0023] The input of the water gauge positioning algorithm is the video image taken by the mobile terminal and the location of the water gauge;

[0024] The output of the water ruler positioning algorithm is the water ruler area image and the water ruler object set A (a linked list character object set, the attributes of the character object include text area, area image, text ellipse attribute, left-associated character object, right-associated character object, upper-associated character object, lower-associated character object, character recognition accuracy, character recognition value, and adjacent object distance);

[0025] This algorithm firstly uses edge detection technology to determine the hull area, secondly, applies the regional fast search algorithm to mark the interest area that meets the color, size, and aspect ratio (the character supplementary search in the rusted area is based on the known water surface boundary line), thirdly, measures the elliptical attributes of the characters in the interest area to preliminarily determine the size, tilt, distortion and other characteristics of the characters, fourthly, is based on the learning framework of convolutional neural network and the recognition of printed characters and character library, and fifthly, performs correlation detection on the identified interest area. Firstly, the characters are arranged in order, and multiple water ruler object sets (linked character object sets) are formed with reference to the water ruler similarity matrix weight table M (size ratio between adjacent characters, adjacent spacing, character tilt, horizontal ratio of adjacent character spacing to character size, vertical ratio of adjacent character spacing to character size, horizontal ratio of adjacent characters, and vertical ratio of adjacent characters) to calculate the familiarity of water ruler objects that meet the water ruler features, and outputs the water ruler object set A with the highest familiarity.

[0026] The water gauge positioning of the present invention is a key technology for ship water gauge measurement. The algorithm solves the water gauge positioning problem under the conditions of severe tilt and distortion of ship water gauge video images and free shooting at a video shooting distance of 1-20 meters, providing a basis for water gauge measurement.

[0027] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in this field.

[0028] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. Although embodiments of the present invention have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and purpose of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for positioning a water gauge of a ship, characterized in that: The following steps are involved: Step 1, applying edge detection technology to determine the hull area; Step 2, applying a fast area search algorithm to mark the areas of interest that meet the color, size, and aspect ratio; Step 3, measuring the ellipse attributes of the characters in the interest area to preliminarily determine the size, tilt, and distortion characteristics of the characters; Step 4: Learning framework based on convolutional neural network and recognition of printed books and character libraries; Step 5, performing correlation detection on the identified interest regions, firstly arranging the characters in order, referring to the water ruler similarity matrix weight table, forming multiple water ruler object sets, calculating the familiarity of the water ruler objects with the water ruler features, and outputting the water ruler object set with the highest familiarity; The water ruler object set is a linked list character object set, and the attributes of the character objects include text area, area image, text ellipse attribute, left-related character object, right-related character object, upper-related character object, lower-related character object, character recognition accuracy, character recognition value, and adjacent object distance; The weight table of the water ruler similarity matrix includes the size ratio between adjacent characters, adjacent spacing, character inclination, the horizontal ratio of adjacent character spacing to character size, the vertical ratio of adjacent character spacing to character size, the horizontal ratio of adjacent characters, and the vertical ratio of adjacent characters.

2. A method for positioning a water gauge of a ship according to claim 1, characterized in that: The input of the water gauge positioning algorithm is the video image taken by the mobile terminal and the location of the water gauge; the output of the algorithm is the water gauge area image and the water gauge object set.

3. A method for positioning a water gauge of a ship according to claim 2, characterized in that: The convolutional neural network structure includes three layers: convolution layer one, the first pooling layer one and normalization layer one. The convolution kernel size of convolution layer one is 5×5, the step size is 4, and there are 155 convolution kernels in total; convolution layer two and normalization layer two, the convolution kernel size of convolution layer two is 4×4, the step size is 3, and there are 158 convolution kernels in total; pooling layer three and normalization layer three, the convolution kernel size of convolution layer three is 3×3, the step size is 2, and there are 144 convolution kernels in total; the classifier adopts Softmax classification.

Images