CN-121482649-B - Unmanned aerial vehicle thermal imaging visual target detection method for search and rescue tasks

CN121482649BCN 121482649 BCN121482649 BCN 121482649BCN-121482649-B

Abstract

The invention relates to the technical field of target detection, in particular to a search and rescue task-oriented unmanned aerial vehicle thermal imaging visual target detection method which comprises the following steps of acquiring unmanned aerial vehicle multi-frame thermal imaging images, extracting region thermal difference characteristics according to windows, marking non-background regions to generate candidate sets, fitting and reconstructing suspected thermal target contours, analyzing tracks and thermal rate to screen background interference, identifying jump abnormality and positioning gravity centers, and generating target repositioning signals. According to the invention, the thermal anomaly area is judged by constructing the thermal value range and variance index sequence in the image area and combining the temperature baseline difference, the background disturbance comparison is carried out by combining the direction vector of the coordinate track in the multi-frame image and the thermal change parameter, the false detection probability caused by background noise is reduced, the target jump recognition is carried out according to the inter-frame heat value and the area change rate linkage thermal contour closure degree, the target discrimination accuracy in the shielding scene is improved, and the robustness of thermal target extraction in the complex search and rescue environment is integrally improved.

Inventors

Wei Yunsu
LI ZHAOKANG
WANG LIN
WANG YANYU
WANG XUEJIAO
GAO YUANZHE
HUANG XUSHENG

Assignees

河北工业职业技术大学

Dates

Publication Date: 20260512
Application Date: 20251124

Claims (7)

1. The unmanned aerial vehicle thermal imaging visual target detection method for search and rescue tasks is characterized by comprising the following steps of: S1, acquiring multi-frame thermal imaging images acquired by an unmanned aerial vehicle in a search and rescue task area, marking image subareas with non-background characteristics according to the degree of deviation of thermal differences from background thermal equilibrium baseline temperature differences, and generating a high-heat candidate positioning area set; S2, judging whether the interval heat value meets fitting conditions or not based on the high heat candidate positioning region set, if so, performing linear gradient fitting processing on the pixel heat value in the interval, reconstructing a suspected human body contour region with continuous boundaries in the region, and generating a suspected target heat morphology region; S3, based on the space position of each outline area in the suspected target thermal morphology area, acquiring a coordinate change track in continuous image frames, comparing the coordinate change track with a path record in a constructed thermal background disturbance track table, screening out background disturbance hot spot areas, and generating an effective thermal target candidate area set; S4, calling the effective thermal target candidate region set, respectively calculating the heat value change rate and the pixel area change rate of the region between frames, judging whether a local shielding or target switching situation exists, acquiring the thermal contour closure degree of the corresponding region in the frames as an auxiliary index, and generating an abnormal track target identification mark set; S5, based on the heat target area marked as jump in the abnormal track target identification mark set, calling the position and heat response distribution of the area in the current frame, extracting hot spot dense center points, performing weighted center-of-gravity positioning processing, constructing an intra-frame area rescanning window, and sending an area normal image, a thermal imaging image and area positioning information to rescue workers to generate a search and rescue target repositioning trigger signal; The search and rescue target repositioning trigger signal acquiring step specifically comprises the following steps: S511, based on the region marked as jump in the abnormal track target identification mark set, extracting a thermal response matrix of a corresponding region in the current image frame, searching pixel points with heat value larger than the median heat value in the region, calculating the barycentric coordinates of the hot spot, giving a weighted value to the heat value of the hot spot, and carrying out two-dimensional weighted average to obtain a region thermal intensity focus point to obtain a thermal response weighted barycentric coordinate set; S512, calling the thermal response weighted barycentric coordinate set, constructing a square area window with the side length twice as long as the original target outline by taking each barycentric coordinate as the center, intercepting an infrared thermal imaging image, a visible light image and a heat value data frame corresponding to the window, constructing an intra-frame target rescanning information structure with the space heat density as an index, and acquiring an area rescanning window data block set; S513, writing the corresponding infrared heat map, visible light image and gravity center positioning coordinates into a search and rescue intermediate interactive frame structure according to the regional rescanning window data block set, synchronously transmitting the infrared heat map, the visible light image and the gravity center positioning coordinates to a front-end rescue node as target information to be confirmed, encoding target data and registering a state field, and establishing a search and rescue target repositioning trigger signal; the search and rescue target repositioning trigger signal comprises a thermal response gravity center coordinate point, an area rescanning window parameter, a trigger condition judgment value and a target detection module activation instruction.
2. The search and rescue mission oriented unmanned aerial vehicle thermal imaging visual target detection method according to claim 1, wherein the high heat candidate positioning region set comprises a multi-window thermal difference significant block, a region thermal response standard deviation value, a thermal difference deviation background reference value mapping matrix and abnormal thermal difference region boundary coordinates, the suspected target thermal form region comprises a thermal value gradient continuous image block, a linear fit thermal profile distribution, a bilinear interpolation complement pixel image layer and a potential target form closed boundary in the fit block, the effective thermal target candidate region set comprises a thermal stability region track set, a spatial target position set after disturbance elimination, thermal response continuous characteristic data and a non-background matching identification coordinate point, and the abnormal track target identification mark set comprises a thermal value jump identification mark, a shielding behavior inference mark, an area mutation region number and a form closure degree judgment mark.
3. The search and rescue task-oriented unmanned aerial vehicle thermal imaging visual target detection method according to claim 2, wherein the step of obtaining the high-heat candidate positioning region set is specifically as follows: S111, acquiring multi-frame thermal imaging images acquired by an unmanned aerial vehicle in a search and rescue task area, dividing each frame of image into window areas with fixed sizes by adopting a regular grid, extracting pixel difference values between maximum heat value pixels in each window and minimum heat value pixels in corresponding peripheral windows to obtain thermal response differences in the window areas, and acquiring a thermal difference response set between the windows; s112, based on the inter-window thermal difference response set, performing range and variance calculation on pixel difference sequences in each image window, establishing a difference statistical index, performing offset comparison on the statistical value in the window area and the reference temperature difference of the whole background of the thermal imaging image, and calculating to obtain a thermal difference response deviation value of the image window area; And S113, comparing the thermal difference response deviation value with a thermal response deviation threshold, screening window numbers exceeding the thermal response deviation threshold, extracting boundary coordinate information of a corresponding image region, carrying out region number and structure classification processing, uniformly marking a region set as a non-background response region, and obtaining a high-heat candidate positioning region set.
4. The search and rescue task-oriented unmanned aerial vehicle thermal imaging visual target detection method according to claim 3, wherein the step of acquiring the suspected target thermal morphology region is specifically as follows: S211, respectively extracting a maximum pixel point and a minimum pixel point of a heat value in each region based on the high heat candidate positioning region set, constructing a heat value region boundary, judging each region boundary according to a set human body surface temperature reference range, screening image regions of which the heat value region falls into the human body surface temperature reference range, and obtaining a target region set conforming to the heat reference region; S212, calling the target area set conforming to the thermal reference interval, carrying out coordinate mapping and numerical extraction on effective pixel points with heat values recorded in each image area, constructing a two-dimensional coordinate and heat value pair set, respectively calculating heat value gradient differences between adjacent pixels in the transverse direction and the longitudinal direction, estimating continuous trend of pixel response in the area according to a linear increment rule, constructing a heat value distribution function curved surface, and generating thermal gradient fitting distribution information; S213, identifying a pixel heat value cavity area caused by missing, shielding or noise suppression in a fitting curved surface according to the thermal gradient fitting distribution information, executing bilinear interpolation on the neighborhood around the cavity, filling the heat value missing, extracting a closed boundary curve in the filled image area according to the heat value change gradient direction, constructing a contour structure diagram, and obtaining a suspected target heat morphology area.
5. The search and rescue task-oriented unmanned aerial vehicle thermal imaging visual target detection method according to claim 4, wherein the step of obtaining the effective thermal target candidate region set is specifically as follows: s311, extracting positioning point coordinates of each target in continuous image frames based on geometric boundary coordinates of each contour structure in the suspected target thermal morphology area, constructing a space position change track in a frame sequence, recording pixel displacement directions and amplitudes between adjacent frames, calculating multi-frame path vectors, and obtaining a contour track direction parameter set; S312, calling the profile track direction parameter set, counting the occurrence times of each coordinate point by combining the corresponding frame number, calculating the number of frames of the targets in the image sequence, extracting a thermal response value change sequence in the corresponding time period, calculating the average temperature change rate according to the inter-frame temperature increase and decrease trend, constructing continuous thermal dynamic characteristics bound with each target profile, and obtaining target thermal time change information; S313, normalizing three characteristic parameters of a track direction vector, a continuous frame number and a temperature change rate according to the target thermal time change information to construct a three-dimensional characteristic vector, carrying out Euclidean distance matching judgment on the three-dimensional characteristic vector and a disturbance path set in a pre-stored thermal background disturbance track table, screening out a background disturbance item with similarity higher than a similarity threshold value, and obtaining an effective thermal target candidate region set.
6. The search and rescue task oriented unmanned aerial vehicle thermal imaging visual target detection method according to claim 5, wherein the normalization processing of the three characteristic parameters of the track direction vector, the continuous frame number and the temperature change rate is specifically implemented by representing the track direction vector in each image frame as a two-dimensional coordinate differential sequence, carrying out proportional normalization on the continuous frame number according to the total frame number, and carrying out amplitude normalization processing on the temperature change rate according to the pixel calorific value change mean value in unit time; The similarity threshold is a distance boundary value calculated based on Euclidean distance mean value and standard deviation of a historical disturbance path in a thermal background disturbance track table in a training stage, and the distance boundary value is the standard deviation of a mean weighting set proportion.
7. The search and rescue task-oriented unmanned aerial vehicle thermal imaging visual target detection method according to claim 6, wherein the step of acquiring the abnormal track target identification mark set is specifically as follows: S411, based on the effective heat target candidate region set, calling the maximum heat value, the minimum heat value and the corresponding frame number information recorded in continuous frames of each region, calculating the heat value difference between the front frame and the rear frame, carrying out normalization processing on the time span of the region, calculating the temperature change amplitude in unit time, and obtaining a heat value inter-frame change rate set; S412, calling the heat value inter-frame change rate set, counting the area change value between continuous frames by combining the number of pixel points of the corresponding region in each image frame, calculating to obtain the jump trend quantity of the thermal target region, comparing the jump trend quantity with the shielding jump critical threshold value, and if the jump trend quantity of the thermal target region is larger than the shielding jump critical threshold value, marking to meet the abnormal change standard to generate a jump sensitive region number set; S413, according to the jump sensitive area number set, searching the thermal contour line closing state of the corresponding area in the image, executing boundary integrity judgment on each thermal contour line, extracting the curve closing rate as a structural stability index, marking the structural boundary unstable area, giving an identification state code, and establishing an abnormal track target identification mark set.

Description

Unmanned aerial vehicle thermal imaging visual target detection method for search and rescue tasks Technical Field The invention relates to the technical field of target detection, in particular to a search and rescue task-oriented unmanned aerial vehicle thermal imaging visual target detection method. Background The field of object detection technology aims to accurately identify and locate multiple objects of interest in an image from the image or video sequence by means of computer vision methods. The target detection field relies on core algorithm frameworks such as deep learning, convolutional neural network, feature pyramid network, region suggestion network and the like, and comprises processing flows such as image input, feature extraction, candidate region generation, target classification, bounding box regression and the like. The target detection is widely applied to a plurality of scenes such as security monitoring, automatic driving, medical image analysis, industrial detection and the like, and particularly, higher requirements are put forward on the detection precision, the real-time performance of the model and the robustness under a complex background. The research focus in the field also comprises the optimization of the adaptability to complex scenes such as shielding, scale change, illumination change, multi-target overlapping and the like, and the deployment and stability of the model performance in practical application are promoted. The unmanned aerial vehicle thermal imaging visual target detection method facing the search and rescue task aims at combining an unmanned aerial vehicle platform and thermal imaging vision to construct a detection method capable of executing target recognition and positioning in a complex environment. The method is mainly used for automatically detecting and calibrating target positions of personnel, animals or other life bodies based on thermal imaging images in emergency search and rescue scenes such as natural disasters, sudden accidents and the like, providing real-time and efficient information support for rescue deployment, improving search and rescue efficiency and accuracy, and reducing search and rescue delay and risk. The traditional detection method relies on a temperature mutation area in a thermal imaging image to conduct static image analysis, and lacks a dynamic quantification means for the difference between the heat value of a local area and the background temperature distribution, so that false marks are easy to appear when the background temperature difference distribution is uneven or a high-temperature interference source exists, a complete heat value change sequence model is not established to identify the continuity of target features in continuous frames, so that target form jump caused by shielding or posture change cannot be effectively identified and confirmed, missing pixel information cannot be compensated in an image edge or resolution reduction scene, and therefore, contour boundary breakage and target identification are incomplete, and stable tracking and effective positioning efficiency of a thermal target are affected. . Disclosure of Invention The invention aims to solve the defects in the prior art, and provides an unmanned aerial vehicle thermal imaging visual target detection method for search and rescue tasks. In order to achieve the purpose, the invention adopts the following technical scheme that the unmanned aerial vehicle thermal imaging visual target detection method for search and rescue tasks comprises the following steps: S1, acquiring multi-frame thermal imaging images acquired by an unmanned aerial vehicle in a search and rescue task area, marking image subareas with non-background characteristics according to the degree of deviation of thermal differences from background thermal equilibrium baseline temperature differences, and generating a high-heat candidate positioning area set; S2, judging whether the interval heat value meets fitting conditions or not based on the high heat candidate positioning region set, if so, performing linear gradient fitting processing on the pixel heat value in the interval, reconstructing a suspected human body contour region with continuous boundaries in the region, and generating a suspected target heat morphology region; S3, based on the space position of each outline area in the suspected target thermal morphology area, acquiring a coordinate change track in continuous image frames, comparing the coordinate change track with a path record in a constructed thermal background disturbance track table, screening out background disturbance hot spot areas, and generating an effective thermal target candidate area set; S4, calling the effective thermal target candidate region set, respectively calculating the heat value change rate and the pixel area change rate of the region between frames, judging whether a local shielding or target switching situation exists, acquiring the th