In-smoke person posture recognition system for fire-fighting simulation training

Abstract

The invention relates to the field of personnel posture recognition, is used for solving the problem that a coach cannot distinguish position information and posture information of a trainee due to low accuracy of personnel action recognition under the condition of smoke shielding in fire-fighting simulation training, and particularly relates to an in-smoke personnel posture recognition system for fire-fighting simulation training. Comprising an image acquisition unit, a smoke identification unit, a personnel identification unit, an action construction unit and an automatic early warning unit. According to the invention, image acquisition is carried out on a smoke-containing area in fire-fighting simulation training, personnel attitude information of the acquired image is subjected to preliminary extraction and identification through a defogging algorithm, and then the smoke area is graded to different degrees; and secondary confirmation is performed on the personnel attitude information extracted and identified from the fuzzy picture according to the smog shielding condition, and the picture and the clear picture under the smog shielding condition are comprehensively analyzed by using the AI technology, so that the visual presentation of the personnel attitude information in the smog is realized.

Description

Technical Field

[0001] The invention relates to the field of personnel posture recognition, in particular to a personnel posture recognition system in smoke for fire simulation training. Background Art

[0002] Real smoke and real fire training is an important way for firefighters to improve their firefighting and rescue capabilities. It allows firefighters to directly feel the changes in temperature, smoke and jet patterns. With the help of wearing air breathing apparatus and pressure control and adjustment of fire trucks, it can improve the combat capabilities of individual firefighters and teams. In firefighter training, fire smoke and heat shroud training is generally used. However, under severe smoke shroud conditions, coaches cannot distinguish the position and posture information of trainees. Therefore, in order to improve the training effect and ensure the safety of trainees, the path information, personnel information, posture information, etc. of the training team inside the smoke and heat cage must be tracked at any time during the training.

[0003] The existing patent CN117057388A discloses a technical solution, which analyzes the echo reception of the radio frequency tag, and thus determines the position and action information of the trainee according to the position of the blocked radio frequency tag. However, in actual training, in order to improve the cooperation ability of the trainees, team collaborative training is required. When multiple personnel are in the same space, the blocked radio frequency tags are numerous and large in area, forming an obscured whole, which makes it impossible to distinguish the action of a certain trainee, making the personnel action recognition less intuitive than the visual image. In the prior art, most visual recognitions tend to study general scenes, and there are fewer studies on personnel detection in special scenes under smoke and occlusion. For the personnel detection methods in smoke occlusion scenes, most of them are to defog first and then detect. Smoke is a non-rigid gas with a phenomenon of uneven dynamic changes. When the smoke is processed by the defog method, the image clarity after processing is not high, resulting in low accuracy of the recognition result.

[0004] In view of the above technical problems, this application proposes a solution. Summary of the invention

[0005] In the present invention, by collecting images of areas containing smoke during fire simulation training, multiple analyses are performed on the collected images to obtain blurred images of personnel positions in the smoke area and smoke occlusion information, and AI technology is used to perform comprehensive analysis on images under smoke occlusion and clear images, thereby achieving restoration of the complete image, and further achieving intuitive representation of the posture information of personnel in the smoke, thereby solving the problem of low accuracy in personnel action recognition under smoke occlusion in fire simulation training, which makes it impossible for coaches to distinguish between the position information and posture information of trainees, and proposes a personnel posture recognition system in smoke for fire simulation training.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A person posture recognition system in smoke for fire simulation training includes an image acquisition unit, which is used to acquire images of a fire simulation training area to obtain an initial image;

[0008] A smoke recognition unit, which can acquire an initial image, perform region segmentation analysis on the initial image, and generate a smoke coverage mark and a smoke concentration classification mark according to the analysis result;

[0009] A person identification unit, after acquiring an initial image, performs algorithmic identification on the initial image to acquire a plurality of human body image regions, compares the human body image regions with preset model parameters, marks different regions as comparison passed or comparison failed according to the comparison results, and generates a comparison image;

[0010] An action construction unit, wherein the action construction unit obtains a comparison image, a smoke coverage mark, and a smoke concentration classification mark, and the action construction unit obtains a personnel action model by combining the comparison image with the smoke coverage mark and the smoke concentration classification mark;

[0011] The automatic warning unit obtains the personnel action model through the action construction unit, compares the personnel action model, and generates a normal posture signal or an abnormal posture signal.

[0012] As a preferred embodiment of the present invention, the smoke recognition unit divides the initial image into i regions, performs algorithm recognition analysis on each of the segmented regions, obtains the area ratio of the smoke in each region, and compares the area ratio of the smoke in the i-th region with a preset ratio threshold. If the area ratio of the smoke is greater than the preset ratio threshold, the i-th region is recorded as a smoke-covered area; otherwise, the i-th region is recorded as a smoke-uncovered area.

[0013] As a preferred embodiment of the present invention, the smoke recognition unit analyzes the distribution concentration of smoke in each segmented area according to the color recognition algorithm, and compares the smoke concentration with the preset concentration gradient values ​​S1, S2, and S3. If the smoke concentration is less than the preset concentration gradient value S1, a smoke-free area is generated; if the smoke concentration is greater than the preset concentration gradient value S3, a heavy smoke area is generated; if the smoke concentration is between the concentration gradient values ​​S1 and S2, a small amount of smoke area is generated; if the smoke concentration is between the concentration gradient values ​​S2 and S3, a medium smoke area is generated;

[0014] The smoke recognition unit sends the segmented area image to the action construction unit, and the area image carries a smoke coverage mark and a smoke concentration classification mark.

[0015] As a preferred embodiment of the present invention, the personnel identification unit removes the smoke area on the initial image by a defogging algorithm, and then performs personnel feature recognition on the removed image. The method by which the personnel identification unit recognizes personnel features is:

[0016] The personnel identification unit acquires a three-dimensional stereoscopic image, and then identifies image parts similar to the human body in the three-dimensional stereoscopic image through an algorithm. The personnel identification unit segments the image parts similar to the human body in the image to obtain multiple human body picture areas. The personnel identification unit performs nested comparison between human body model parameters and a human body picture area, and records different areas as comparison passed or comparison failed according to the comparison results. The personnel identification unit combines all comparison passed areas and comparison failed areas to generate a comparison image.

[0017] As a preferred implementation of the present invention, after acquiring the comparison image, the action construction unit overlaps the comparison image with the segmented region image to generate the action construction image.

[0018] As a preferred implementation of the present invention, the action construction unit generates a construction image in the following manner:

[0019] If both the smoke-uncovered area and the smoke-free area are comparison-passed areas, the comparison-passed area will be confirmed; if both are comparison-failed areas, the comparison-failed area will be confirmed;

[0020] Mark the area where the smoke covered area and the heavy smoke area overlap on the contrast image and record it as a fuzzy area. If the fuzzy area is also a contrast passed area, the contrast passed area is cancelled. If the fuzzy area is also a contrast failed area, the contrast failed area is confirmed as the area to be generated.

[0021] Mark the area where the smoke covered area and the medium smoke area overlap on the contrast image and record it as a slightly blurred area. If the slightly blurred area is also a contrast passed area, the contrast passed area is confirmed. If the slightly blurred area is also a contrast failed area, the contrast failed area is confirmed as the area to be generated.

[0022] The action construction unit counts all confirmed areas and generates a personnel posture construction screen.

[0023] As a preferred embodiment of the present invention, the method for the action construction unit to generate a personnel posture construction screen is:

[0024] The confirmed comparisons are combined by regions to generate a person posture image with missing parts;

[0025] Fill the confirmed areas to be generated and the areas where comparison failed to be completed into the corresponding positions in the person posture image with missing parts;

[0026] The AI ​​algorithm combines the comparison passing area and the comparison failing area to complete the image of the area to be generated, forming a complete image of the person's posture.

[0027] The action construction unit performs personnel recognition on the personnel posture construction picture, obtains the personnel position, and captures the personnel position through an algorithm to obtain the personnel action model.

[0028] As a preferred embodiment of the present invention, the action construction unit sends the personnel action model to the automatic warning unit, the automatic warning unit compares the personnel action model with multiple preset standard action models, calculates the overlap between the personnel action model and the preset standard action model, selects the group with the largest overlap as the overlap matching result, and compares the overlap matching result with the preset threshold value. If the overlap matching result is less than the preset threshold value, an abnormal posture signal is generated, otherwise a normal posture signal is generated.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. In the present invention, in the fire simulation training, by collecting images of the area containing smoke, multiple analyses are performed on the collected images to obtain the blurred image of the position of people in the smoke area and the smoke occlusion information, and AI technology is used to perform a comprehensive analysis of the picture under smoke occlusion and the clear picture, so as to realize the restoration of the complete picture, and then realize the intuitive expression of the posture information of people in the smoke.

[0031] 2. In the present invention, the defogging algorithm is used to perform preliminary extraction and identification of the posture information of the person in the blurred image, and then the smoke area is classified to different degrees, and the posture information of the person extracted and identified in the blurred image is reconfirmed according to the smoke occlusion situation, so as to determine the feasibility of image restoration and the credibility of the available clear image, thereby improving the accuracy of image restoration.

[0032] 3. In the present invention, the identified personnel posture information is automatically identified and judged, so that an alarm reminder can be automatically generated when the personnel posture information is abnormal, which expands the functionality of the personnel posture recognition system and improves the intelligence level of fire simulation training. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0034] Figure 1 is a system block diagram of the present invention;

[0035] Figure 2 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Embodiment 1:

[0038] See also Figure 1 - Figure 2 As shown, a person posture recognition system in smoke for fire simulation training includes an image acquisition unit, a smoke recognition unit, a person recognition unit, an action construction unit and an automatic warning unit. The image acquisition unit acquires images of the fire simulation training area through a camera device, records the acquired images as initial images, and sends the initial images to the smoke recognition unit and the person recognition unit respectively.

[0039] After obtaining the initial image, the personnel identification unit removes the smoke area on the initial image through the defogging algorithm, and then performs personnel feature recognition on the removed image. The recognized personnel features include the torso area image, the limb area image and the face area image. The method for the personnel identification unit to recognize personnel features is as follows:

[0040] Step 1: The person recognition unit applies a color recognition algorithm to the defogging area, obtains a three-dimensional image based on the color recognition algorithm and the spatial depth algorithm, and then recognizes the image parts similar to the human body in the three-dimensional image through the algorithm;

[0041] Step 2: The person recognition unit segments the image parts similar to the human body in the image to obtain multiple human body image areas;

[0042] Step 3: compare multiple human image regions one by one with preset human model parameters, the human model parameters include a human torso model, a human limb model and a human face model, the person recognition unit nests the human model parameters with a human image region, if a model in the human model parameters and a human image region have a nesting matching rate greater than a preset threshold, it is recorded as a comparison pass, and the human model parameter part is marked for the human image region that has passed the comparison, if the human model parameters and a human image region have a nesting matching rate less than the preset threshold, it is recorded as a comparison failure;

[0043] Step 4: The person recognition unit combines all the contrast passing areas and contrast failing areas to generate a contrast image, and sends the contrast image to the action construction unit.

[0044] Embodiment 2:

[0045] See also Figure 1 - Figure 2 As shown, after the smoke recognition unit obtains the initial image through the image acquisition unit, it divides the initial image into regions, thereby dividing the initial image into i regions, performing algorithm recognition analysis on each of the segmented regions, obtaining the area ratio of smoke in each region, and comparing the area ratio of smoke in the i-th region with a preset ratio threshold. If the area ratio of smoke is greater than the preset ratio threshold, the i-th region is recorded as a smoke covered region. If the area ratio of smoke is less than the preset ratio threshold, the i-th region is recorded as a smoke uncovered region.

[0046] The smoke recognition unit analyzes the distribution density of smoke in each segmented area according to the color recognition algorithm to obtain the smoke density. The smoke recognition unit compares the smoke density with the preset concentration gradient values ​​S1, S2, and S3. If the smoke density is less than the preset concentration gradient value S1, a smoke-free area is generated. If the smoke density is greater than the preset concentration gradient value S3, a heavy smoke area is generated. If the smoke density is between the concentration gradient values ​​S1 and S2, a small amount of smoke area is generated. If the smoke density is between the concentration gradient values ​​S2 and S3, a medium smoke area is generated.

[0047] The smoke recognition unit sends the segmented area image to the action construction unit. The area image carries a smoke coverage mark and a smoke concentration classification mark. The smoke concentration coverage mark is divided into a smoke covered area or a smoke covered area. The smoke concentration classification mark includes a smoke-free area, a small amount of smoke area, a medium smoke area and a heavy smoke area.

[0048] Embodiment three:

[0049] See also Figure 1 - Figure 2 As shown, after the action construction unit obtains the comparison image, it overlaps the comparison image with the segmented region image to generate the action construction image. The action construction unit generates the construction image in the following manner:

[0050] Mark the area not covered by smoke, the area without smoke and the area with a small amount of smoke on the comparison image. If the area not covered by smoke and the area without smoke are both the areas passed by the comparison, the area passed by the comparison will be confirmed. If they are both the areas failed by the comparison, the area failed by the comparison will be confirmed.

[0051] Mark the area where the smoke covered area and the heavy smoke area overlap on the contrast image and record it as a fuzzy area. If the fuzzy area is also a contrast passed area, the contrast passed area is cancelled. If the fuzzy area is also a contrast failed area, the contrast failed area is confirmed as the area to be generated.

[0052] Mark the area where the smoke covered area and the medium smoke area overlap on the contrast image and record it as a slightly blurred area. If the slightly blurred area is also a contrast passed area, the contrast passed area is confirmed. If the slightly blurred area is also a contrast failed area, the contrast failed area is confirmed as the area to be generated.

[0053] The action construction unit counts all confirmed areas and generates a personnel posture construction screen. The method for generating the personnel posture construction screen is:

[0054] The confirmed comparisons are combined by regions to generate a person posture image with missing parts;

[0055] Fill the confirmed area to be generated into the corresponding position in the person posture picture with the missing part;

[0056] Fill in the area confirmed as failed comparison to the corresponding position in the person posture image with the missing part;

[0057] The AI ​​algorithm combines the comparison passing area and the comparison failing area to complete the image of the area to be generated, forming a complete image of the person's posture.

[0058] The action construction unit performs personnel recognition on the personnel posture construction screen, obtains the personnel position, and captures the personnel position through an algorithm to obtain the personnel action model;

[0059] The action construction unit sends the personnel action model to the automatic warning unit, which compares the personnel action model with a plurality of preset standard action models, calculates the overlap between the personnel action model and the preset standard action models, selects the group with the largest overlap as the overlap matching result, and compares the overlap matching result with the preset threshold value. If the overlap matching result is greater than or equal to the preset threshold value, a normal posture signal is generated; if the overlap matching result is less than the preset threshold value, an abnormal posture signal is generated.

[0060] After generating the abnormal posture signal, the automatic warning unit issues an alarm reminder through the sound and light alarm device, and at the same time outputs the action model of the person who generated the abnormal posture signal through the display device.

[0061] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A smoke-based personnel posture recognition system for fire simulation training, characterized in that: It includes an image acquisition unit, which is used to acquire images of the fire simulation training area to obtain an initial image; A smoke recognition unit, which can acquire an initial image, perform region segmentation analysis on the initial image, and generate a smoke coverage mark and a smoke concentration classification mark according to the analysis result; A person identification unit, after acquiring an initial image, performs algorithmic identification on the initial image to acquire a plurality of human body image regions, compares the human body image regions with preset model parameters, marks different regions as comparison passed or comparison failed according to the comparison results, and generates a comparison image; An action construction unit, wherein the action construction unit obtains a comparison image, a smoke coverage mark, and a smoke concentration classification mark, and the action construction unit obtains a personnel action model by combining the comparison image with the smoke coverage mark and the smoke concentration classification mark; The automatic warning unit obtains the personnel motion model through the motion construction unit, and compares the personnel motion model to generate a normal posture signal or an abnormal posture signal.

2. The smoke-based personnel posture recognition system for fire simulation training according to claim 1 is characterized in that: The smoke recognition unit divides the initial image into i regions, performs algorithm recognition analysis on each of the segmented regions, obtains the area ratio of the smoke in each region, and compares the area ratio of the smoke in the i-th region with a preset ratio threshold. If the area ratio of the smoke is greater than the preset ratio threshold, the i-th region is recorded as a smoke-covered area; otherwise, the i-th region is recorded as a smoke-uncovered area.

3. The smoke-based personnel posture recognition system for fire simulation training according to claim 2 is characterized in that: The smoke recognition unit analyzes the distribution density of smoke in each segmented area according to the color recognition algorithm, and compares the smoke density with the preset concentration gradient values ​​S1, S2, and S3. If the smoke density is less than the preset concentration gradient value S1, a smoke-free area is generated; if the smoke density is greater than the preset concentration gradient value S3, a heavy smoke area is generated; if the smoke density is between the concentration gradient values ​​S1 and S2, a small amount of smoke area is generated; if the smoke density is between the concentration gradient values ​​S2 and S3, a medium smoke area is generated; The smoke recognition unit sends the segmented area image to the action construction unit, and the area image carries a smoke coverage mark and a smoke concentration classification mark.

4. The smoke-based personnel posture recognition system for fire simulation training according to claim 1 is characterized in that: The personnel identification unit removes the smoke area on the initial image by using a defogging algorithm, and then performs personnel feature recognition on the removed image. The method by which the personnel identification unit recognizes personnel features is: The personnel identification unit acquires a three-dimensional stereoscopic image, and then identifies image parts similar to the human body in the three-dimensional stereoscopic image through an algorithm. The personnel identification unit segments the image parts similar to the human body in the image to obtain multiple human body picture areas. The personnel identification unit performs nested comparison between human body model parameters and a human body picture area, and records different areas as comparison passed or comparison failed according to the comparison results. The personnel identification unit combines all comparison passed areas and comparison failed areas to generate a comparison image.

5. The smoke-based personnel posture recognition system for fire simulation training according to claim 1 is characterized in that: After acquiring the comparison image, the action construction unit overlaps the comparison image with the segmented region image to generate the action construction image.

6. The smoke-based personnel posture recognition system for fire simulation training according to claim 5 is characterized in that: The action construction unit generates a construction image in the following manner: If both the smoke-uncovered area and the smoke-free area are comparison-passed areas, the comparison-passed area will be confirmed; if both are comparison-failed areas, the comparison-failed area will be confirmed; Mark the area where the smoke covered area and the heavy smoke area overlap on the contrast image and record it as a fuzzy area. If the fuzzy area is also a contrast passed area, the contrast passed area is cancelled. If the fuzzy area is also a contrast failed area, the contrast failed area is confirmed as the area to be generated. Mark the area where the smoke covered area and the medium smoke area overlap on the contrast image and record it as a slightly blurred area. If the slightly blurred area is also a contrast passed area, the contrast passed area is confirmed. If the slightly blurred area is also a contrast failed area, the contrast failed area is confirmed as the area to be generated. The action construction unit counts all confirmed areas and generates a personnel posture construction screen.

7. The smoke-based personnel posture recognition system for fire simulation training according to claim 6 is characterized in that: The method for the action construction unit to generate a personnel posture construction picture is: The confirmed comparisons are combined by regions to generate a person posture image with missing parts; Fill the confirmed areas to be generated and the areas where comparison failed to be completed into the corresponding positions in the person posture image with the missing parts; The AI ​​algorithm combines the comparison passing area and the comparison failing area to complete the image of the area to be generated, forming a complete image of the person's posture. The action construction unit performs personnel recognition on the personnel posture construction picture, obtains the personnel position, and captures the personnel position through an algorithm to obtain the personnel action model.

8. The smoke-based personnel posture recognition system for fire simulation training according to claim 1 is characterized in that: The action construction unit sends the personnel action model to the automatic warning unit, which compares the personnel action model with multiple preset standard action models, calculates the degree of overlap between the personnel action model and the preset standard action models, selects the group with the largest overlap as the overlap matching result, and compares the overlap matching result with the preset threshold. If the overlap matching result is less than the preset threshold, an abnormal posture signal is generated, otherwise a normal posture signal is generated.

Images