CN-116863172-B - Automobile collision early warning icon identification method

CN116863172BCN 116863172 BCN116863172 BCN 116863172BCN-116863172-B

Abstract

According to the method, firstly, an inter-frame similarity sequence is calculated through a color histogram, so that a key frame image is obtained, then, through feature extraction and matching of a selected reference image, rectangular frame coordinates of an early warning icon area are obtained after perspective transformation, then, slicing processing is carried out on other key frame images according to the rectangular frame coordinates of the early warning icon area, effective matching point pairs of all key frame images are obtained respectively through feature extraction and matching of sliced areas of each key frame image, the identified image containing the early warning icon can be selected after the comparison with the preset reference number, recognition of the automobile collision early warning icon is achieved, key frames containing time information can be output, the automobile collision early warning icon can be identified rapidly and effectively, the testing efficiency of an automobile collision system is effectively improved, and the consumption of manpower resources is reduced.

Inventors

LI ZHONGLI
SUN JINGGE
CHEN GUANGXIU
WANG JUN
HAN CHONG
WANG SHUAI
MA LIXIANG

Assignees

河南科技大学

Dates

Publication Date: 20260505
Application Date: 20230704

Claims (2)

1. The method for identifying the automobile collision early warning icon is characterized by comprising the following steps of: Firstly, shooting an obstacle target on a running path in front of a vehicle in real time through a vehicle-mounted camera device, obtaining the real-time distance between the vehicle and the obstacle target through GPS positioning, when the real-time distance is within a preset distance range, reading frame-by-frame images of a real-time shot video of a vehicle instrument panel, sequentially adding time stamps to each frame of read image, obtaining an image set to be identified which is ordered according to the time stamps, and respectively drawing color histograms of all images in the image set to be identified, so as to obtain a color histogram set which is ordered according to the time stamps; Step two, reading a color histogram set, and sequentially calculating the similarity of the color histogram between the color histogram of each time stamp and the color histogram of the next adjacent time stamp to obtain a color histogram similarity sequence ordered according to the time stamps; the calculation formula of the similarity of the color histogram is as follows: ; In the above-mentioned method, the step of, The i-th point of the color histogram curve that is the first time stamp, The ith point of the color histogram curve that is the next time stamp; Sequentially performing convolution smoothing on all color histogram similarities in the color histogram similarity sequence through a hanning window to obtain an inter-frame similarity sequence ordered according to a time stamp, respectively comparing each inter-frame similarity in the inter-frame similarity sequence with the adjacent previous and next inter-frame similarities, and selecting an image in an image set to be identified corresponding to the inter-frame similarity as a key frame image when the certain inter-frame similarity is simultaneously smaller than the adjacent two inter-frame similarities in the sequence to obtain a key frame image set; Selecting a reference image containing an early warning icon from the last ten frames of images in the key frame image set according to the time stamp sequence, cutting the early warning icon from the reference image, and respectively performing SIFT feature extraction on the reference image and the early warning icon to respectively obtain a feature point group of the reference image and a feature point group of the early warning icon; Fifth, matching the feature point group of the reference image and the feature point group of the early warning icon through a FLANN matching algorithm to obtain nearest neighbor matching distance and next neighbor matching distance of each feature point, selecting the feature point corresponding to the feature point and the nearest neighbor matching distance as a matching point pair when the ratio of the nearest neighbor matching distance to the next neighbor matching distance is smaller than a threshold value, removing the wrong matching point pair through a RANSAC random sampling algorithm, reserving at least four effective matching point pairs, and then carrying out homography transformation on the effective matching point pairs to obtain a homography matrix; Step six, performing perspective transformation on the early warning icon and the homography matrix to obtain rectangular frame coordinates of the early warning icon area; Step seven, slicing the keyframe images except the reference image according to the rectangular frame coordinates of the early warning icon area to obtain an early warning icon rectangular area of the keyframe image, and performing SIFT feature extraction on the early warning icon rectangular area to obtain feature point groups of all the keyframe images except the reference image; Step eight, respectively matching the characteristic point group of the early warning icon and the characteristic point groups of the plurality of key frame images through a FLANN matching algorithm to obtain the nearest matching distance and the next adjacent matching distance of each characteristic point, and selecting the characteristic point corresponding to the characteristic point and the nearest matching distance as a matching point pair when the ratio of the nearest matching distance to the next adjacent matching distance is smaller than a threshold value to obtain the effective matching point pair number respectively corresponding to the plurality of key frame images; And step nine, comparing the number of the effective matching points obtained in the step eight with the preset reference number respectively, and selecting the key frame image corresponding to the number of the effective matching points as the identified image containing the early warning icon when the number of the effective matching points is larger than the preset reference number.
2. The method for identifying an icon for warning of an automobile collision according to claim 1, wherein in the fifth step, the homography matrix is a matrix of 3x3, expressed as: 。

Description

Automobile collision early warning icon identification method Technical Field The invention relates to the field of automobile collision early warning, in particular to an automobile collision early warning icon recognition method. Background Currently, automobile manufacturers and parts suppliers worldwide have begun to develop and commercialize forward collision warning systems for vehicles, which have been gradually moved to the market. Correspondingly, the international standardization organization issues ISO 15623:2013 'performance requirements and test procedures of the intelligent transportation system forward collision warning system', and according to the standard and the national standard GB/T20608-2006 'performance requirements and detection methods of the intelligent transportation system self-adaptive cruise control system', the national standard GB/T33577-2017 'performance requirements and test procedures of the intelligent transportation system vehicle forward collision warning system' are issued in China, and the GB/T33577-2017 standard specifies the system functions, requirements and performance test procedures of the collision warning system. The performance test comprises the test of a detection area, the test of alarm distance and precision and the test of target identification capability, wherein each test is further divided into a plurality of tests according to the performance requirement, and repeated experiments are needed. In the prior art, the video of the instrument panel recorded by the camera during the test is generally searched manually to find out the moment when the early warning icon appears, and then the data analysis is carried out by combining the two-vehicle GPS information and the vehicle speed information recorded by other instruments at the moment. However, in view of the fact that such a test requires repeated experiments, the method of using manual recognition is very inefficient, and a specific image of the occurrence of the car collision warning icon and the occurrence position thereof cannot be obtained accurately before the test is performed, so that an unknown icon needs to be recognized. If the early warning icons are acquired in advance to learn, a large amount of image training and machine learning time are needed in advance to achieve the purpose of real-time detection, and the appearance forms and positions of the automobile collision early warning icons of different automobile manufacturers are different, so that the practicability is poor. Disclosure of Invention The invention aims to provide an automobile collision early warning icon identification method, which effectively improves the efficiency of automobile collision system test and reduces the consumption of human resources. The technical scheme adopted by the invention for solving the technical problems is that the method for identifying the automobile collision early warning icon comprises the following steps: Firstly, shooting an obstacle target on a running path in front of a vehicle in real time through a vehicle-mounted camera device, obtaining the real-time distance between the vehicle and the obstacle target through GPS positioning, when the real-time distance is within a preset distance range, reading frame-by-frame images of a real-time shot video of a vehicle instrument panel, sequentially adding time stamps to each frame of read image, obtaining an image set to be identified which is ordered according to the time stamps, and respectively drawing color histograms of all images in the image set to be identified, so as to obtain a color histogram set which is ordered according to the time stamps; Step two, reading a color histogram set, and sequentially calculating the similarity of the color histogram between the color histogram of each time stamp and the color histogram of the next adjacent time stamp to obtain a color histogram similarity sequence ordered according to the time stamps; the calculation formula of the similarity of the color histogram is as follows: ; In the above-mentioned method, the step of, The i-th point of the color histogram curve that is the first time stamp,The ith point of the color histogram curve that is the next time stamp; Sequentially performing convolution smoothing on all color histogram similarities in the color histogram similarity sequence through a hanning window to obtain an inter-frame similarity sequence ordered according to a time stamp, respectively comparing each inter-frame similarity in the inter-frame similarity sequence with the adjacent previous and next inter-frame similarities, and selecting an image in an image set to be identified corresponding to the inter-frame similarity as a key frame image when the certain inter-frame similarity is simultaneously smaller than the adjacent two inter-frame similarities in the sequence to obtain a key frame image set; Selecting a reference image containing an early warning icon from the last ten fram