CN-121985225-A - Moving target inhibition method for long exposure monitoring imaging and security camera

CN121985225ACN 121985225 ACN121985225 ACN 121985225ACN-121985225-A

Abstract

The invention discloses a moving target inhibition method for long exposure monitoring imaging and a security camera, and relates to the technical field of image processing; the method comprises the following steps of multi-exposure image acquisition, synchronous acquisition of a frame of long exposure image and a short exposure image sequence of the same scene, moving target detection and track analysis, detection of a moving target in the scene and analysis of the moving track of the moving target based on the short exposure image sequence, self-adaptive exposure control, and dynamic adjustment of exposure parameters of subsequent image acquisition according to scene brightness and the moving state of the moving target. The invention fundamentally solves the contradiction between the lifting brightness of long exposure and the blurring of moving targets by synchronously collecting two-way heterogeneous images, the finally output image approaches to the long exposure image in the whole brightness and dynamic range, and the edge and detail definition of the moving targets approach to the short exposure image, thereby obviously inhibiting motion smear and blurring and improving the discernability of valuable targets in low-illumination monitoring scenes.

Inventors

WANG CHEN
QIAN HAIQIANG
HUANG TAO

Assignees

苏州维卓智能科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260316

Claims (10)

1. The method for inhibiting the moving target of the long exposure monitoring imaging is characterized by comprising the following steps of: s1, multi-exposure image acquisition, namely synchronously acquiring a frame of long exposure image and a short exposure image sequence of the same scene; s2, detecting a moving target and analyzing a track, and detecting the moving target in a scene and analyzing the moving track based on the short exposure image sequence; s3, self-adaptive exposure control, wherein exposure parameters of subsequent image acquisition are dynamically adjusted according to scene brightness and the motion state of the moving object; S4, image fusion and reconstruction, wherein the long exposure image and the short exposure image sequence are fused to generate a fusion image; and S5, motion artifact suppression, namely identifying and suppressing blurring or smear artifact generated by target motion in the fused image.
2. The method for suppressing a moving object for long-exposure monitoring imaging according to claim 1, wherein the multi-exposure image acquisition specifically comprises: S11, dividing incident light to a main image sensor and an auxiliary image sensor through a light splitting element; S12, controlling the main image sensor to acquire and obtain the long exposure image with a first exposure time; And S13, controlling the auxiliary image sensor to acquire the short exposure image sequence at a second exposure time, wherein the second exposure time is shorter than the first exposure time, and acquiring a plurality of frames in the first exposure time period.
3. The method for suppressing a moving object for long-exposure monitoring imaging according to claim 2, wherein the moving object detection and trajectory analysis specifically comprises: s21, preprocessing the short exposure image sequence to perform time domain noise reduction and contrast enhancement; s22, calculating a dense optical flow field between adjacent short exposure frames to obtain a pixel-level motion vector; s23, dividing a motion foreground region based on the consistency of the amplitude and the direction of the motion vector, and clustering the motion foreground region into independent motion targets; s24, carrying out multi-frame tracking on each moving target to form a moving track of each moving target, and predicting a moving path of each moving target in the first exposure time.
4. The method for suppressing a moving object for long-exposure monitoring imaging according to claim 3, wherein in said adaptive exposure control, according to ambient illuminance And intensity of exercise Dynamically calculating the long exposure time for the next acquisition The calculation formula is as follows: Wherein, the As a function of the base exposure time, For the target brightness to be the same, In order to adjust the sensitivity factor for the brightness, In order for the motion suppression intensity factor to be of a magnitude, Is a comprehensive index calculated based on the maximum speed of the moving object and the total area of the moving area.
5. The method for suppressing a moving object for long-exposure monitoring imaging according to claim 3, wherein the image fusion and reconstruction specifically comprises: S41, selecting a frame from the short exposure image sequence as a reference clear frame; s42, using the dense light flow field to align the background areas of the long exposure image and other frames in the short exposure image sequence with the reference clear frame; s43, generating a fusion weight map according to the moving target area information, wherein a moving area gives higher weight to the reference clear frame, and a static area gives higher weight to the long exposure image; s44, inputting the aligned long exposure image, the reference clear frame and the weight map into a trained deep learning fusion network to generate an initial fusion image.
6. The method for suppressing a moving object for long exposure monitoring imaging according to claim 5, wherein the motion artifact suppression specifically comprises: S51, combining the predicted motion path of the moving target, and generating an accurate motion artifact mask, wherein the mask identifies a fuzzy smear region caused by long exposure in the initial fusion image; S52, taking the motion artifact mask as a guide, adopting a context attention repair network, extracting texture information from a static background area of the initial fusion image, and repairing the fuzzy smear area; and S53, performing color consistency correction and self-adaptive sharpening on the repaired image, and outputting a final image.
7. The method for suppressing a moving object in long exposure monitoring imaging according to claim 5, wherein the deep learning fusion network is a U-Net architecture, and a loss function is combined with an L1 loss function and a multi-scale structure similarity loss function.
8. A security camera, which comprises a camera body and a camera body, characterized by comprising the following steps: The image acquisition module is used for synchronously acquiring a frame of long exposure image and a short exposure image sequence of the same scene; the motion analysis module is used for detecting a moving target based on the short exposure image sequence and analyzing the motion trail of the moving target; The exposure control module is used for dynamically adjusting exposure parameters according to scene brightness and motion states; the image processing module is used for fusing the long exposure image with the short exposure image sequence and inhibiting motion artifacts in the fused image; the storage and transmission module is used for storing and outputting the processed image data; The image processing module includes: An image registration unit for aligning the multi-frame images; The motion compensation unit is used for compensating the image area according to the motion trail output by the motion analysis module; An image fusion unit, built-in deep learning accelerator, for running the deep learning fusion network of claim 7.
9. The security camera of claim 8, wherein the image acquisition module comprises: the beam splitting prism is used for splitting incident light into two paths; The main image sensor is arranged on the first light path, and the pixel size of the main image sensor is larger than a first threshold value and is used for acquiring the long exposure image; and the auxiliary image sensor is arranged on the second light path, and the frame rate reading capacity of the auxiliary image sensor is higher than a second threshold value and is used for acquiring the short exposure image sequence.
10. The security camera of claim 8, wherein the image processing module comprises an artifact suppression unit configured to perform the motion artifact suppression step of claim 6, comprising generating a motion artifact mask and running a contextual attention repair network.

Description

Moving target inhibition method for long exposure monitoring imaging and security camera Technical Field The invention relates to the technical field of image processing, in particular to a moving target inhibition method for long-exposure monitoring imaging and a security camera. Background In the fields of security monitoring, intelligent traffic, night observation and the like, cameras are often required to work in low-illumination environments. In order to obtain a sufficiently bright image, extending the exposure time is a straightforward and common technical approach. By increasing the photosensitive time of the photoelectric sensor, the signal-to-noise ratio and the overall brightness of the image can be remarkably improved. However, this technique has an inherent, insurmountable disadvantage in that the relative motion of any object in the scene during the exposure time causes displacement in the imaging plane, resulting in blurring, smearing or even complete blurring of the moving object. The method seriously influences the subsequent feature recognition, behavior analysis and track tracking of moving targets (such as pedestrians and vehicles), and greatly restricts the application value of the long exposure technology in dynamic monitoring scenes. To alleviate the motion blur problem, the prior art mainly starts from the image processing level and proposes various post-processing solutions, such as: The method is based on image registration and averaging, and the method tries to estimate global or local motion between frames by collecting continuous multi-frame short exposure images and using an optical flow method or feature matching, and then performs fusion or averaging after aligning the sequence images to the same reference coordinate system. However, such methods have high computational complexity, poor real-time performance, and easy error in the estimation of fast, complex movements, leading to registration failure, and "ghosting" or warping in the fused image. The method is based on the fusion of motion detection and selectivity, and firstly, static areas and dynamic areas (moving targets) in an image sequence are distinguished through background modeling or a frame difference method. At fusion, the clear information is extracted mainly from a single frame or static region, while the fuzzy data from multiple frames of the dynamic region is attempted to be abandoned or weakened. The method has the main defects that the accuracy of motion detection is extremely easy to be interfered by illumination change, shadow and noise, and the method essentially sacrifices the texture details of a moving target, so that the information of a target area is lost, and a clear moving object image cannot be recovered. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a moving target inhibition method for long exposure monitoring imaging and a security camera. In order to achieve the above purpose, the present invention adopts the following technical scheme: a moving target inhibition method for long exposure monitoring imaging comprises the following steps: s1, multi-exposure image acquisition, namely synchronously acquiring a frame of long exposure image and a short exposure image sequence of the same scene; s2, detecting a moving target and analyzing a track, and detecting the moving target in a scene and analyzing the moving track based on the short exposure image sequence; s3, self-adaptive exposure control, wherein exposure parameters of subsequent image acquisition are dynamically adjusted according to scene brightness and the motion state of the moving object; S4, image fusion and reconstruction, wherein the long exposure image and the short exposure image sequence are fused to generate a fusion image; and S5, motion artifact suppression, namely identifying and suppressing blurring or smear artifact generated by target motion in the fused image. Preferably, the multi-exposure image acquisition specifically comprises: S11, dividing incident light to a main image sensor and an auxiliary image sensor through a light splitting element; S12, controlling the main image sensor to acquire and obtain the long exposure image with a first exposure time; And S13, controlling the auxiliary image sensor to acquire the short exposure image sequence at a second exposure time, wherein the second exposure time is shorter than the first exposure time, and acquiring a plurality of frames in the first exposure time period. Preferably, the moving object detection and track analysis specifically comprises: s21, preprocessing the short exposure image sequence to perform time domain noise reduction and contrast enhancement; s22, calculating a dense optical flow field between adjacent short exposure frames to obtain a pixel-level motion vector; s23, dividing a motion foreground region based on the consistency of the amplitude and the direction of the motion vector, and clustering t