CN-121747081-B - Vehicle-mounted thermal imaging target AI identification method supporting aerial imaging display

Abstract

The invention relates to the technical field of image data processing, in particular to a vehicle-mounted thermal imaging target AI identification method supporting aerial imaging display. The method comprises the steps of obtaining thermal imaging video stream and vehicle speed data shot in the running process of a vehicle, identifying each suspected pedestrian region of adjacent image frames in the thermal imaging video stream, calculating the region similarity of each corresponding suspected pedestrian region of the adjacent image frames, carrying out picture consistency matching on the adjacent image frames to enable the adjacent image frames to finish image alignment processing, reflecting the change trend of an integral picture by using the vehicle speed data, combining the region similarity of each suspected pedestrian region in an alignment picture to obtain a pedestrian body region and a pedestrian limb region in the alignment picture, confirming whether pedestrian features exist in the alignment picture truly or not, and carrying out marking display on the corresponding regions when the pedestrian features exist in the corresponding regions. The technical scheme of the invention can realize accurate pedestrian recognition.

Inventors

• LUO JINTAO

Assignees

• 长沙市鑫泰仪器有限公司

Dates

Publication Date

20260512

Application Date

20260302

Claims (5)

1. 1. An AI-recognition method for a vehicle-mounted thermal imaging target supporting aerial imaging display, the method comprising: Acquiring a thermal imaging video stream shot in the running process of a vehicle and vehicle speed data corresponding to a shooting period; Identifying each suspected pedestrian region of adjacent image frames in the thermal imaging video stream, and determining the region similarity of each corresponding suspected pedestrian region of the adjacent image frames; According to the similarity of all the areas and the distribution characteristics of each suspected pedestrian area in the image frame, carrying out picture consistency matching on the adjacent image frames so as to enable the adjacent image frames to finish image alignment processing; extracting characteristic regions according to the vehicle speed data and the region similarity of the suspected pedestrian region in the alignment picture to obtain a pedestrian trunk region and a pedestrian limb region in the alignment picture; Analyzing whether the pedestrian body area and the pedestrian limb area have pedestrian characteristics, and displaying corresponding pedestrian areas by marking when the pedestrian characteristics are determined to exist; the method for determining the regional similarity comprises the following steps: Image segmentation is carried out on adjacent image frames of the thermal imaging video stream according to a preset gray threshold value, so that an area formed by pixel points larger than the gray threshold value in the adjacent image frames is determined to be a suspected pedestrian area; Calculating the centroid of the irregular graph of the suspected pedestrian area, performing overlapping comparison, and counting the number of pixels in the overlapping area and the non-overlapping area to obtain the number of first pixels in the overlapping area and the number of second pixels in the non-overlapping area when the positions of different centroids overlap; summing the centroid positions of each suspected pedestrian region when the different centroid positions are overlapped with the distance of each pixel point in the overlapped region, and taking the sum as region distance data corresponding to the overlapped result; obtaining a temperature difference coefficient corresponding to an overlapping result according to the number of first pixel points, the area distance data, the minimum circumcircle radius of an overlapping area and the gray level difference value of each pixel point of the overlapping area, wherein the first pixel points are overlapped at different centroid positions of the adjacent image frames; Obtaining the regional similarity of each corresponding suspected pedestrian region of the adjacent image frames according to the temperature difference coefficient, the first pixel point number and the second pixel point number of each group of suspected pedestrian regions; The method for determining the temperature difference coefficient comprises the following steps: Obtaining single-point deviation degree of each pixel point and the centroid position according to the difference between the minimum circumcircle radius of the overlapping region and the distance value of the corresponding pixel point in the region distance data when the adjacent image frames are overlapped at different centroid positions; obtaining the area deviation degree of all pixel points in the overlapped area from the centroid position according to the distance summation result of the first pixel point number, the minimum circumscribing circle radius and the area distance data, wherein the calculation formula is as follows , For the distance summation result of the corresponding region distance data, Representing the radius of the smallest circumscribing circle, Representing the number of first pixel points; according to all single-point deviation degrees, area deviation degrees and gray level difference values of all pixel points of each group of suspected pedestrian areas, obtaining temperature difference coefficients of each group of suspected pedestrian areas, and calculating the temperature difference coefficients as follows: In the formula (I), in the formula (II), The coefficient of temperature difference is represented by a graph, A distance value representing the i-th pixel point, Indicating the single point deviation degree of the ith pixel point and the centroid position, Representing the gray difference value of each pixel point in the overlapped area; the method for completing image alignment processing by adjacent image frames comprises the following steps: Obtaining picture matching degrees of picture consistency matching of the adjacent image frames by different suspected pedestrian areas according to the similarity of all the areas and the distribution characteristics of each suspected pedestrian area in the image frames; determining a suspected pedestrian area corresponding to the maximum picture matching degree of all picture matching degrees as a target pedestrian area; determining the centroid position of the target pedestrian area as an alignment reference of the adjacent image frames, and overlapping and aligning the two image frames; the method for determining the pedestrian trunk area and the pedestrian limb area comprises the following steps: Determining an area with the area similarity larger than a similarity threshold value of the suspected pedestrian area in the alignment picture as a pedestrian body area; obtaining a speed influence coefficient of a pedestrian body area in an alignment picture according to a preset maximum value and an average value of the vehicle speed data, wherein the corresponding formula is as follows: In which, in the process, Represents a preset maximum value of the vehicle speed data, Mean value of the vehicle speed data; According to the regional similarity, the average gray level difference value and the speed influence coefficient of the pedestrian trunk region, a gray level evaluation threshold value of the pedestrian limb region around the pedestrian trunk region is obtained, and the corresponding formula is as follows: In the formula (I), in the formula (II), The gray scale evaluation threshold value is represented, Representing the average gray level difference of the pedestrian trunk area, The region similarity representing the pedestrian trunk region, Is a security item; And when the gray level difference value of the pixels of the adjacent image frames positioned at the periphery of the pedestrian body area is larger than the gray level evaluation threshold value, determining that the corresponding pixels are pixels of the pedestrian body area.
2. 2. The method for identifying an AI target by vehicle thermal imaging for supporting aerial imaging display according to claim 1, wherein the obtaining the region similarity of each corresponding suspected pedestrian region of the adjacent image frames according to the temperature difference coefficient, the first pixel number and the second pixel number of each group of suspected pedestrian regions comprises: Obtaining an overlapping duty ratio coefficient when the corresponding centroid positions are overlapped according to the ratio of the number of the first pixel points to the number of the second pixel points when the different centroid positions are overlapped; according to the overlapping duty ratio coefficient and the temperature difference coefficient when the centroid positions are overlapped, the regional similarity of each corresponding suspected pedestrian region of the adjacent image frames is obtained, and the corresponding formula is as follows: In the formula (I), in the formula (II), Representing the similarity of the two suspected pedestrian areas, Representing the number of first pixels, Representing the number of the second pixel points, Representing the normalization of the maximum and minimum values, Representing the temperature difference coefficient.
3. 3. The method for identifying an AI-based thermal imaging target supporting aerial imaging display according to claim 1, wherein the obtaining a picture matching degree of picture consistency matching of the adjacent image frames with different suspected pedestrian areas according to all the area similarities and distribution characteristics of each suspected pedestrian area in the image frames comprises: The similarity of all the areas is arranged in a descending order, and the suspected pedestrian areas corresponding to the similarity of all the areas are sequentially selected according to the ordering result to perform preliminary overlapping alignment of pictures; Obtaining the area quantity ratio of the number of the picture areas corresponding to the suspected pedestrian areas in the overlapped alignment picture to the total number of all the suspected pedestrian areas in the adjacent image frames according to the preliminary overlapped alignment result of the pictures of the different suspected pedestrian areas; And obtaining the picture matching degree of the adjacent image frames in the preliminary overlapping alignment of different pictures according to the number proportion of the regions in the preliminary overlapping alignment of the pictures, the extremely poor position distance in the alignment regions and the similarity average value of the similarity of all the regions, wherein the corresponding formulas are as follows: In the formula (I), in the formula (II), The degree of matching of the picture is indicated, To align picture area The similarity mean of the similarity of all the regions in the region, The total number of suspected pedestrian areas is X for the minimum distance of positions in the alignment area.
4. 4. The method for identifying an on-board thermal imaging target AI supporting an aerial imaging display of claim 1, wherein said analyzing whether pedestrian features are present in the pedestrian trunk area and the pedestrian extremity area comprises: Respectively carrying out pixel statistics on the pedestrian trunk area and the pedestrian limb area to obtain the third pixel number of each pedestrian trunk area and the fourth pixel number of the corresponding peripheral pedestrian limb area; Obtaining the pedestrian existence probability of the corresponding suspected pedestrian area according to the area similarity of the pedestrian body area of the adjacent image frames, the similarity mean value of all the suspected pedestrian areas in the aligned picture, the third pixel point number and the fourth pixel point number; when the pedestrian existence probability is larger than a preset probability threshold value, determining that pedestrian features exist in the corresponding suspected pedestrian area; the method for acquiring the pedestrian existence probability comprises the following steps: Obtaining a similarity coefficient of pedestrians in the corresponding suspected pedestrian area according to the ratio of the area similarity of the pedestrian body area of the adjacent image frames to the average value of the similarities of all the suspected pedestrian areas in the alignment picture; Obtaining a pixel point number coefficient corresponding to the pedestrian in the suspected pedestrian area according to the difference value between the third pixel point number and the fourth pixel point number; Obtaining the pedestrian existence probability of the corresponding suspected pedestrian area according to the similarity coefficient and the pixel point number coefficient, wherein the corresponding formula is that In which, in the process, The regional similarity for the pedestrian's trunk area, In order to align the average value of the similarity of all the suspected pedestrian areas in the picture, the number of pixel points in the pedestrian limb areas is recorded as The number of pixel points in the pedestrian trunk area is recorded as , Is a normalization function.
5. 5. The method for identifying an on-board thermal imaging target AI supporting an aerial imaging display according to claim 1, wherein the marking displays a corresponding pedestrian area when it is determined that a pedestrian feature exists, comprising: holographic floating projection is carried out at a corresponding position of the driver's visual field according to the position of the pedestrian area on the image frame.

Description

Vehicle-mounted thermal imaging target AI identification method supporting aerial imaging display Technical Field The invention relates to the technical field of image data processing, in particular to a vehicle-mounted thermal imaging target AI identification method supporting aerial imaging display. Background With the popularization of the household automobiles, the frequency of self-driving travel is increased, the travel is facilitated, but the occurrence frequency of traffic accidents is further increased, and most traffic accidents occur in bad traffic environments such as night, rainy days, foggy days and the like. Because the visual field of the driver is limited, pedestrians, riders and the like do not have obvious lamplight marks, the driver cannot find the positions of the pedestrians in time, and serious traffic accidents are caused. Therefore, in a poor traffic environment where the driver's visual field is limited, it is important to help the driver to locate the position of the living things around the vehicle in time by using the auxiliary driving system. At present, an infrared night vision driving auxiliary system is mounted on an automobile, and even in a bad traffic environment, the automobile can work in a completely non-light environment through a thermal imaging technology through a vehicle-mounted safety device based on the thermal imaging technology, so that the automobile can still quickly identify the situation of the position of a pedestrian in a large range, provide more reaction time for a driver, and effectively avoid potential hazards such as traffic accidents. When the position of the pedestrian in the infrared image is identified, the pedestrian is always obvious in the infrared thermal imaging image because the body temperature of the human body is constant and a certain temperature difference exists between the pedestrian and the surrounding environment, the pedestrian can be identified and marked directly according to a threshold segmentation algorithm, but the motion area obtained based on a frame difference method is larger, and the pedestrian contour is easy to be divided wrongly. Disclosure of Invention In order to solve the technical problem of accurately dividing the area where pedestrians are located in the vehicle driving process, the invention aims to provide the vehicle-mounted thermal imaging target AI identification method supporting aerial imaging display, and the influence of continuous thermal imaging picture changes on the division of the pedestrian contours in the vehicle driving process is eliminated. The adopted technical scheme is as follows: The embodiment of the invention provides a vehicle-mounted thermal imaging target AI identification method supporting aerial imaging display, which comprises the following steps: Acquiring a thermal imaging video stream shot in the running process of a vehicle and vehicle speed data corresponding to a shooting period; Identifying each suspected pedestrian area of adjacent image frames in the thermal imaging video stream, and determining the area similarity of each corresponding suspected pedestrian area of the adjacent image frames; according to the similarity of all the areas and the distribution characteristics of each suspected pedestrian area in the image frame, carrying out picture consistency matching on the adjacent image frames so as to enable the adjacent image frames to finish image alignment processing; Extracting characteristic regions according to the vehicle speed data and the region similarity of the suspected pedestrian region in the alignment picture to obtain a pedestrian trunk region and a pedestrian limb region in the alignment picture; and analyzing whether pedestrian characteristics exist in the pedestrian trunk area and the pedestrian limb area, and displaying corresponding pedestrian areas in a marked mode when the existence of the pedestrian characteristics is determined. In an alternative embodiment, identifying each suspected pedestrian region of adjacent image frames in the thermal imaging video stream and determining a region similarity of each corresponding suspected pedestrian region of the adjacent image frames comprises: Image segmentation is carried out on adjacent image frames of the thermal imaging video stream according to a preset gray threshold value, so that an area formed by pixel points larger than the gray threshold value in the adjacent image frames is determined to be a suspected pedestrian area; Calculating the centroid of the irregular graph of the suspected pedestrian area, performing overlapping comparison, and counting the number of pixels in the overlapping area and the non-overlapping area to obtain the number of first pixels in the overlapping area and the number of second pixels in the non-overlapping area when overlapping at different centroid positions; obtaining region distance data corresponding to an overlapping result according to the distances between