A monitoring image enhancement processing method in low-illumination environment

Abstract

The application discloses a kind of low-illumination environment monitoring image enhancement processing method, it is related to video monitoring technical field, comprising: obtaining input monitoring image, determining dominant light source area, and extracting highlight coverage distribution information and edge visible distribution information along dominant light source area outward, according to the continuous coverage termination position in highlight coverage distribution information and the continuous visible termination position in edge visible distribution information determine the obscuration intrusion boundary;Extract monitoring target candidate structure that is intersected with obscuration intrusion boundary or located inside obscuration intrusion boundary, and monitoring target candidate structure is divided into light side structure, dark side structure and intermediate transition structure according to the incident direction of dominant light source.The application first establishes the determination process of obscuration intrusion boundary around dominant light source area, and the starting range in local strong light source neighborhood is identified by the combination analysis of highlight coverage distribution information and edge visible distribution information.

Description

Technical Field

[0001] This invention relates to the field of video surveillance technology, specifically to a method for enhancing surveillance images in low-light environments. Background Technology

[0002] In existing surveillance applications, image enhancement in low-light environments has always been an important research direction in the field of image processing. For scenarios such as nighttime, underground spaces, park roads, and entrances and exits, existing technologies typically improve image quality through methods such as brightness enhancement, contrast enhancement, noise suppression, local area enhancement, defogging, or edge enhancement, thereby improving the visibility of the surveillance footage and the usability of subsequent target recognition. These technologies have good application value in general low-light scenarios and have formed a relatively rich set of implementation methods. Meanwhile, in some nighttime surveillance scenarios, there are also local strong light sources such as streetlights, supplementary lights, and vehicle lights in the monitored area. When the air contains fog, moisture, or fine water vapor, a continuously expanding bright coverage area is easily formed around the local strong light source, affecting the edge visibility of nearby monitored targets. At this time, image degradation often does not occur uniformly across the entire image, but is concentrated in the neighborhood of the dominant light source, and often manifests as the bright coverage area encroaching on one side of the target structure, causing changes in the target's edge, outline, or local closure relationship. Although existing image enhancement schemes can improve low-light or foggy images as a whole, more targeted processing methods are usually still needed for this directional invasion feature in the neighborhood of local strong light sources. Furthermore, in the aforementioned scenarios, the monitored target is not necessarily entirely invisible; rather, it may exhibit a situation where the structure on the light-facing side is compromised, the structure on the back-lit side still retains some visible information, and the middle area is in a transitional state. For such monitoring images, simply applying uniform brightness compensation or local enhancement is often insufficient to simultaneously address the release processing of the compromised area, the zoning determination of the target structure, and the subsequent closure and restoration requirements. Therefore, how to construct an enhancement processing flow for monitoring and interpreting damaged structures based on information such as the high-brightness coverage distribution, edge visibility distribution, compromise depth, and structural closure benchmark in the neighborhood of a local strong light source has become a worthy research direction in low-light monitoring image processing. Consequently, this invention proposes a monitoring image enhancement processing method for low-light environments. Summary of the Invention

[0003] The purpose of this invention is to provide a method for enhancing surveillance images in low-light environments to solve the problems mentioned in the background art.

[0004] This invention can be achieved through the following technical solution: a method for enhancing surveillance images in low-light environments, comprising: Step 1: Acquire the input monitoring image, determine the dominant light source area, and extract the highlight coverage distribution information and edge visible distribution information outward from the dominant light source area. Determine the occlusion intrusion boundary based on the continuous coverage termination position in the highlight coverage distribution information and the continuous visible termination position in the edge visible distribution information. Step 2: Extract the candidate structures of the monitoring targets that intersect with or are located inside the shielding intrusion boundary. Divide the candidate structures of the monitoring targets into the light-facing side structure, the back-light side structure, and the intermediate transition structure according to the incident direction of the dominant light source. Determine the intrusion depth based on the intersection position of the shielding intrusion boundary and the light-facing side structure, as well as the termination position of the continuous edge missing state or the termination position of the continuous closed interruption state in the light-facing side structure. Step 3: The candidate structures of the monitoring targets whose invasion depth reaches the preset conditions are identified as damaged structures for monitoring and interpretation. Based on the invasion depth, the light-facing side structure of the damaged structure is shielded and released to obtain the light-facing side restored structure. Step 4: Extract the residual structural segments of the backlight side structure from the damaged structure in the monitoring and interpretation, determine the structural closure benchmark based on the residual structural segments, and perform closure and restoration processing on the missing structural segments of the intermediate transition structure in the light-facing restored structure and the damaged structure in the monitoring and interpretation based on the structural closure benchmark to obtain the monitoring and interpretation restored structure. Step 5: Perform a closure determination on the monitored and restored structure. If the closure determination result does not meet the preset closure conditions, adjust the occlusion release process in Step 3 and the closure restoration process in Step 4. If the closure determination result meets the preset closure conditions, output the enhanced image.

[0005] A further technical improvement of the present invention is that the method for determining the shielding intrusion boundary in step one includes: The bright connected regions in the input monitoring image are extracted, and the bright connected regions with the largest brightness peak and continuous outward bright coverage along multiple preset directions are identified as the dominant light source regions. Starting from the sampling position where the brightness peak is located in the dominant light source region, the high brightness coverage state and edge visibility state of each sampling position are extracted sequentially outward along multiple preset directions to generate high brightness coverage distribution information and edge visibility distribution information corresponding to each preset direction. In each preset direction, the outermost sampling position that is continuously in a highlighted coverage state is determined as the continuous coverage termination position, and the innermost sampling position that is continuously in an edge-visible state is determined as the continuous visibility termination position. Within the segment between the continuous coverage termination position and the continuous visibility termination position, the first sampling position where the edge visibility state begins to change from continuous visibility to continuous weakening while the corresponding highlight coverage state remains continuous is determined as the occlusion intrusion position in the preset direction. When the difference in radial distance between the shielding intrusion positions in adjacent preset directions is less than a preset distance threshold, it is determined that the shielding intrusion positions in adjacent preset directions meet the continuous distribution condition. By sequentially connecting the shielding intrusion positions in adjacent preset directions that meet the continuous distribution condition, the shielding intrusion boundary is obtained.

[0006] A further technical improvement of the present invention is that the method for determining the incident direction of the dominant light source in step two includes the following steps: By performing a localization analysis on the brightness distribution within the dominant light source region, the location of the brightness peak within the dominant light source region is obtained; By analyzing the location of the boundary between the candidate structure of the monitoring target and the boundary of the intrusion, the boundary position closest to the brightness peak position is obtained as the contact position on the light-facing side. Starting from the contact position on the light-facing side, continuously missing edge response segments are detected point by point into the candidate structure of the monitored target along the direction from the peak brightness position to the contact position on the light-facing side. The termination position of the continuously missing edge response segment is obtained as the intrusion extension position. On the boundary of the candidate structure of the monitoring target, starting from the affected extension position, search for continuous visible edge segments along the side away from the brightness peak position, and determine the starting position of the first continuous visible edge segment obtained by the search as the backlight side reserved position; A first directional basic quantity is formed from the brightness peak position to the contact position on the light-facing side, and a second directional basic quantity is formed from the contact position on the light-facing side to the intruded extension position; When the directional deviation between the first directional base quantity and the second directional base quantity is less than a preset directional deviation threshold, the second directional base quantity is used as the dominant light source incident direction. When the directional deviation between the first directional base quantity and the second directional base quantity is greater than or equal to the preset directional deviation threshold, the midpoint of the line connecting the affected extension position and the backlight side retention position is taken as the correction position, and the direction from the brightness peak position to the correction position is taken as the incident direction of the dominant light source.

[0007] A further technical improvement of the present invention is that the method for dividing the candidate structure of the monitoring target in step two includes: The direction of the dominant light source incident is used as the structural dividing direction; the first dividing line is formed along the direction perpendicular to the dominant light source incident direction with the affected extension position as the reference point; and the second dividing line is formed along the direction perpendicular to the dominant light source incident direction with the backlight side retention position as the reference point. The portion of the candidate structure of the monitoring target located on the side facing the dominant light source area of ​​the first dividing line is determined as the light-facing side structure; the portion of the candidate structure of the monitoring target located on the side away from the dominant light source area of ​​the second dividing line is determined as the backlight side structure; and the portion of the candidate structure of the monitoring target located between the first dividing line and the second dividing line is determined as the intermediate transition structure.

[0008] A further technical improvement of the present invention lies in: the method for determining the depth of invasion in step two includes: The boundary between the shielding intrusion boundary and the sun-facing structure is determined as the initiation point of the intrusion. Determine the termination position of the continuous edge missing state in the structure on the sun-facing side; Determine the termination position of the continuous closed interruption state in the structure facing the light; The position that is farther from the start of the invasion in the direction of the dominant light source incident is determined as the invasion termination position, between the termination position of the continuous edge missing state and the termination position of the continuous closed interruption state. The depth of invasion is defined as the length of the extension along the incident direction of the dominant light source between the intrusion start position and the intrusion termination position.

[0009] A further technical improvement of the present invention is that the preset condition in step three is: The depth of the intrusion was compared with the length of the structure on the light-facing side along the incident direction of the dominant light source. When the ratio of the penetration depth to the extension length reaches a preset ratio threshold, the penetration depth is determined to meet the depth condition. Determine the termination positions of continuous edge missing state and continuous closure interruption state in the structure facing the light; when the termination positions of continuous edge missing state and continuous closure interruption state are located in the same affected section, determine that the structure facing the light meets the damage consistency condition. When the depth of the intrusion meets the depth condition and the structure on the sun-facing side meets the damage consistency condition, the candidate structure of the monitoring target is determined as the damaged structure for monitoring and interpretation.

[0010] A further technical improvement of the present invention is that the masking release process in step three includes the following steps: The sampling section along the incident direction of the dominant light source between the intrusion start position and the intrusion termination position is defined as the shielding release section. Starting from the termination point of the invasion, the highlight coverage status of each sampling position in the masking release section is detected point by point along the direction towards the start point of the invasion. The first sampling position where the highlight coverage status changes from continuous to interrupted is determined, and the sampling position located outside the first sampling position and adjacent to it is determined as the outer release boundary. Starting from the initial location of the invasion, the edge visibility status of each sampling location within the occluded release section is detected point by point along the direction toward the termination location of the invasion. The first sampling location where the edge visibility status changes from continuous absence to continuous visibility is determined as the inner release boundary. The sampling location between the outer release boundary and the inner release boundary is determined as the current release segment; The brightness values ​​corresponding to each sampling position within the current release segment are reduced in stages according to the order from the nearest to the starting position of the invasion, with a preset decreasing amplitude. After each stage of reduction, the first sampling position where the edge visibility state changes from continuous missing to continuous visible is re-determined as the updated inner release boundary. When the updated inner release boundary moves to the preset back-off position between the invasion start position and the original inner release boundary, the phased reduction of the brightness value corresponding to each sampling position in the current release segment is stopped, and the processed light-facing side structure is determined as the light-facing side recovery structure.

[0011] A further technical improvement of the present invention lies in the method for determining the structural closure datum in step four, which includes: Determine a continuously visible residual structural segment on the boundary of the backlight-side structure; Following the extension sequence of the residual structural segments, the boundary orientation of each sampling location and the sampling interval of each adjacent sampling location along the extension direction of the residual structural segments are determined sequentially. Using the opposite direction of the incident direction of the dominant light source as the extension direction, the first visible edge position obtained by searching along the extension direction in the intermediate transition structure and the light-facing recovery structure from each sampling position is determined as the corresponding landing point. Based on the boundary orientation change and sampling interval relationship between adjacent sampling positions in the residual structural segment, the relative positions of each corresponding landing point are sequentially adjusted so that the adjusted adjacent corresponding landing points maintain the same orientation change and sampling interval relationship as the residual structural segment. The sequence of corresponding landing points after sequential adjustment is determined as the structural closure benchmark.

[0012] Compared with the prior art, the present invention has the following beneficial effects: This invention first establishes a process for determining the occlusion intrusion boundary around the dominant light source region. By combining and analyzing the distribution information of highlight coverage and the distribution information of visible edges, the starting range of the intrusion in the neighborhood of a local strong light source is identified. This method does not simply enhance the entire image uniformly, but first locates the spatial relationship between the continuous outward expansion of highlight coverage and the weakening of visible edges, thereby more accurately determining the intruded area and its boundary basis, providing support for subsequent structural division and depth determination. Furthermore, this invention achieves a layered characterization of the damaged state of the monitored target by continuously determining the incident direction of the dominant light source, the structure on the light-facing side, the structure on the backlight side, the intermediate transition structure, and the depth of intrusion. Based on this, by determining whether the candidate structure of the monitored target belongs to the damaged structure for monitoring and interpretation through preset conditions, and performing a masking and release process on the structure on the light-facing side, the high brightness coverage of the intruded area in the vicinity of the local strong light source can be gradually retreated, and the edge visibility state can be gradually restored. Thus, the identifiability of the local intruded area can be improved more specifically, rather than just remaining at the level of overall brightness enhancement. On the other hand, this invention utilizes the residual structural segments in the backlight-side structure to determine the structural closure benchmark, and accordingly performs closure restoration processing on the missing structural segments in the front-light-side restored structure and the intermediate transition structure. Then, it adjusts the processing results in conjunction with the closure judgment, making the monitoring and interpretation restored structure more consistent in terms of boundary orientation, sampling interval, and closure relationship. Through this technical chain built around the residual structural segments, the structural closure benchmark, and the closure restoration processing, this invention can restore the structural continuity and closure of the monitoring target even when there is directional intrusion in the neighborhood of a local strong light source, thereby improving the interpretation effect and processing stability of monitoring images in low-light environments. Attached Figure Description

[0013] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0014] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0015] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0016] Please see Figure 1 As shown, the present invention provides a method for enhancing surveillance images in low-light environments, comprising: Step 1: Acquire the input monitoring image, determine the dominant light source area, and extract the highlight coverage distribution information and edge visible distribution information outward from the dominant light source area. Determine the occlusion intrusion boundary based on the continuous coverage termination position in the highlight coverage distribution information and the continuous visible termination position in the edge visible distribution information. Methods for determining the shielded intrusion boundary include: The bright connected regions in the input monitoring image are extracted, and the bright connected regions with the largest brightness peak and continuous outward bright coverage along multiple preset directions are identified as the dominant light source regions. Starting from the sampling position where the brightness peak is located in the dominant light source region, the high brightness coverage state and edge visibility state of each sampling position are extracted sequentially outward along multiple preset directions to generate high brightness coverage distribution information and edge visibility distribution information corresponding to each preset direction. In each preset direction, the outermost sampling position that is continuously in a highlighted coverage state is determined as the continuous coverage termination position, and the innermost sampling position that is continuously in an edge-visible state is determined as the continuous visibility termination position. Within the segment between the continuous coverage termination position and the continuous visibility termination position, the first sampling position where the edge visibility state begins to change from continuous visibility to continuous weakening while the corresponding highlight coverage state remains continuous is determined as the occlusion intrusion position in the preset direction. When the difference in radial distance between the shielding intrusion positions in adjacent preset directions is less than a preset distance threshold, it is determined that the shielding intrusion positions in adjacent preset directions meet the continuous distribution condition. By sequentially connecting the shielding intrusion positions in adjacent preset directions that meet the continuous distribution condition, the shielding intrusion boundary is obtained.

[0017] Specifically, first determine the dominant light source area: After acquiring the input monitoring image, connectivity analysis is performed on the pixel brightness in the input monitoring image. Specifically, the annular region within a preset width range surrounding the area to be determined as the background statistical region is used as the background statistical region. The average pixel brightness within this background statistical region is calculated as the local background brightness average. When the pixel brightness within a candidate pixel region is higher than a preset multiple of the local background brightness average, the candidate pixel region is determined as a bright pixel region. Spatially connected bright pixel regions are merged to form bright connected regions. Subsequently, the brightness peak value is determined for each bright connected region, and the number of sampling points continuously in a bright coverage state is counted point by point along multiple preset directions, starting from the position of the brightness peak value. The product of the number of continuous sampling points and the sampling interval is determined as the continuous outward bright coverage length in the preset direction. When the brightness peak value of a certain bright connected region is the largest, and the number of directions in multiple preset directions that satisfy the continuous outward bright coverage length greater than a preset length threshold reaches the preset number of directions, the bright connected region is determined as the dominant light source region. In one embodiment, the preset width of the background statistical region can be determined based on the equivalent radius of the candidate bright pixel region. In one embodiment, the preset multiplier can be determined based on the degree of difference between the local background brightness distribution and the brightness distribution of the highlight area; In one embodiment, the degree of morphological change of the dominant light source region is determined by analyzing the boundary orientation changes between adjacent boundary sampling positions on the boundary of the dominant light source region; and the number of preset directions is determined based on the degree of morphological change and the requirements of directional sampling for the integrity of the boundary coverage of the dominant light source region.

[0018] In one embodiment, the equivalent size of the dominant light source region is first determined based on the area of ​​the dominant light source region. Then, the number of sampling positions continuously in a high-brightness coverage state is detected point by point outward from the position of the brightness peak along multiple preset directions. The continuous outward high-brightness coverage range is determined based on the statistical results of the continuous high-brightness coverage length in each preset direction. After that, a preset length threshold is determined based on the correspondence between the equivalent size of the dominant light source region and the continuous outward high-brightness coverage range.

[0019] Starting from the sampling position where the brightness peak in the dominant light source region is located, high brightness coverage distribution information and edge visibility distribution information are obtained along multiple preset directions: After determining the dominant light source region, the sampling position where the brightness peak in the dominant light source region is located is used as the starting position for directional sampling, and sampling positions are sequentially set outward along multiple preset directions. For each sampling position, the local brightness value and edge response value of that sampling position are obtained. The local brightness value can be obtained from the average pixel brightness within the neighborhood window of that sampling position, and the edge response value can be obtained from the brightness change amplitude between that sampling position and its adjacent sampling positions. In one embodiment, the larger of the absolute values ​​of the brightness differences between the current sampling position and the adjacent sampling positions along the current preset direction can be determined as the edge response value of that sampling position. When the local brightness value of a sampling position is higher than a preset brightness threshold, and the sampling position and its adjacent sampling positions facing the dominant light source region continuously satisfy the condition that the local brightness value is higher than the preset brightness threshold, the sampling position is determined to be in a high-brightness coverage state. When the edge response value of a sampling position is higher than a preset edge threshold, and the sampling position and its adjacent sampling positions continuously satisfy the condition that the edge response value is higher than the preset edge threshold, the sampling position is determined to be in an edge-visible state. A sequence of sampling positions that continuously satisfy the high-brightness coverage state constitutes the high-brightness coverage distribution information along a corresponding preset direction, and a sequence of sampling positions that continuously satisfy the edge visibility state constitutes the edge visibility distribution information along a corresponding preset direction. The preset brightness threshold can be obtained by multiplying the average brightness of the first ring of sampling positions outside the dominant light source region by 1.2 to 1.5 times, and the preset edge threshold can be obtained by multiplying the median value of the edge response values ​​of all sampling positions in the input monitoring image by 1.1 to 1.4 times. In this embodiment, "continuous" means that at least two consecutive adjacent sampling positions along the same preset direction simultaneously satisfy the corresponding state determination condition. In each preset direction, determine the continuous coverage termination position, the continuous visibility termination position, and the occlusion intrusion position.

[0020] After obtaining the highlight coverage distribution information and edge visibility distribution information in each preset direction, position determination is performed for each preset direction. The outermost sampling position that is continuously in a highlight coverage state along the preset direction is determined as the continuous coverage termination position; the innermost sampling position that is continuously in an edge visibility state along the preset direction is determined as the continuous visibility termination position. Subsequently, in the segment between the continuous coverage termination position and the continuous visibility termination position, the edge response value change is detected point by point in order from the area closer to the dominant light source area to the area farther away from the dominant light source area; when the edge response value of a certain sampling position begins to continuously decrease relative to its previous sampling position, and the number of continuously decreasing sampling positions reaches a preset decrease number, while the sampling position still meets the highlight coverage state determination condition, the sampling position is determined as the occlusion intrusion position in the preset direction. The preset decrease number can be 2 or 3 consecutive sampling positions.

[0021] The shielding intrusion boundary is determined based on the shielding intrusion locations in multiple preset directions: After obtaining the intrusion locations in multiple preset directions, a continuous distribution determination is made for the intrusion locations in adjacent preset directions. Specifically, the radial distance between the intrusion locations in adjacent preset directions and the sampling location of the brightness peak in the dominant light source region is calculated, and the difference between adjacent radial distances is calculated. When the difference is less than a preset distance threshold, the intrusion locations in adjacent preset directions are determined to meet the continuous distribution condition. Subsequently, the intrusion locations in adjacent preset directions that meet the continuous distribution condition are connected in a preset direction order to obtain the intrusion boundary. The preset distance threshold can be 8% to 12% of the average radial distance of multiple intrusion locations, or 5% to 15% of the equivalent radius of the dominant light source region. The intrusion boundary obtained in the above manner is used in the subsequent step two to extract the monitoring target candidate structure that intersects with or is located inside the intrusion boundary, and to further determine the incident direction of the dominant light source, divide the front-side structure, back-side structure and intermediate transition structure, and determine the intrusion depth, thereby maintaining a consistent technical logic with the subsequent embodiments.

[0022] Step 2: Extract the candidate structures of the monitoring targets that intersect with or are located inside the shielding intrusion boundary. Divide the candidate structures of the monitoring targets into the light-facing side structure, the back-light side structure, and the intermediate transition structure according to the incident direction of the dominant light source. Determine the intrusion depth based on the intersection position of the shielding intrusion boundary and the light-facing side structure, as well as the termination position of the continuous edge missing state or the termination position of the continuous closed interruption state in the light-facing side structure. Specifically, based on the dominant light source region and the occlusion intrusion boundary already determined in step one, candidate structures of the monitoring target are obtained and the incident direction of the dominant light source is determined:In this embodiment, candidate structures of the monitoring target are first acquired inside the intrusion shielding boundary. Specifically, edge response values ​​and local brightness values ​​are acquired point by point inside the intrusion shielding boundary, and the distribution of edge response values ​​and local brightness values ​​at each sampling location are statistically analyzed to obtain edge response statistics and brightness statistics. A preset candidate edge threshold is determined based on the average response value of consecutive edge sampling locations in the edge response statistics, and a preset candidate brightness threshold is determined based on the average brightness value of consecutive brightness sampling locations in the brightness statistics. Sampling locations that simultaneously satisfy the condition that the edge response value is higher than the preset candidate edge threshold and the local brightness value is higher than the preset candidate brightness threshold are determined as candidate sampling locations. Subsequently, taking each candidate sampling position as the center, it is checked whether its surrounding adjacent sampling positions also belong to the candidate sampling position. When the sampling distance between the adjacent sampling position and the current candidate sampling position is not greater than the preset adjacency distance threshold, and there is no interruption segment between them where the edge response value is continuously lower than the preset edge missing threshold, the adjacent sampling position and the current candidate sampling position are classified into the same candidate region. The same rule is continued to be applied to newly classified candidate sampling positions for outward expansion until no new candidate sampling positions meet the classification conditions, thus completing the spatial aggregation of a monitoring target candidate structure. The preset adjacency distance threshold is determined based on the average sampling distance between adjacent sampling positions in the input monitoring image. After the monitoring target candidate structure is acquired, the brightness distribution in the dominant light source region is analyzed to determine the brightness peak position in the dominant light source region as the sampling position with the largest brightness value. Then, the boundary position between the monitoring target candidate structure and the occlusion intrusion boundary is analyzed to determine the boundary position closest to the brightness peak position as the contact position on the light-facing side. Starting from the contact position on the light-facing side, edge response values ​​are detected point by point along the direction from the peak brightness position to the contact position on the light-facing side into the candidate structure of the monitored target. A preset edge missing threshold is determined based on the average edge response values ​​in a preset neighborhood around the contact position on the light-facing side. When multiple sampling positions with edge response values ​​lower than the preset edge missing threshold appear consecutively along the direction, and there are no intervening sampling positions where the edge response value is higher than the preset edge missing threshold again, this continuous sampling position sequence is determined as a continuous edge response missing segment, and the sampling position farthest along the direction in this continuous sampling position sequence is determined as the invaded extension position. Subsequently, on the boundary of the candidate structure of the monitored target, starting from the invaded extension position, a boundary sampling position sequence with edge response values ​​continuously higher than the preset edge threshold in step one is searched point by point along the side away from the peak brightness position, and the first sampling position in this boundary sampling position sequence is determined as the backlight-side reserved position.A first directional baseline is formed from the brightness peak position to the contact position on the light-facing side, and a second directional baseline is formed from the contact position on the light-facing side to the intruded extension position. The horizontal and vertical components of the first and second directional baselines in the image coordinates are obtained respectively. The absolute values ​​of the difference between the horizontal and vertical components are added together to obtain the directional deviation value. A preset directional deviation threshold is determined based on the statistical results of the directional deviation values ​​corresponding to multiple candidate structures of the monitored targets. When the directional deviation value is less than the preset directional deviation threshold, the second directional baseline is used as the incident direction of the dominant light source. When the directional deviation value is greater than or equal to the preset directional deviation threshold, the midpoint of the line connecting the intruded extension position and the backlight-retained position is used as the correction position, and the direction from the brightness peak position to the correction position is used as the incident direction of the dominant light source.

[0023] After determining the incident direction of the dominant light source, the structure is divided into the front-light side structure, the back-light side structure, and the intermediate transition structure: After obtaining the incident direction of the dominant light source, the affected extension position, and the reserved position on the backlight side, the incident direction of the dominant light source is used as the structural division direction. Specifically, using the affected extension position as a reference point, a first dividing line is formed by extending the line perpendicular to the incident direction of the dominant light source towards both sides of the candidate structure boundary; using the reserved position on the backlight side as a reference point, a second dividing line is formed by extending the line perpendicular to the incident direction of the dominant light source towards both sides of the candidate structure boundary. Subsequently, the portion of the candidate structure located on the side facing the dominant light source region along the first dividing line is defined as the light-facing structure; the portion located on the side away from the dominant light source region along the second dividing line is defined as the backlight side structure; and the portion of the candidate structure located between the first and second dividing lines is defined as the intermediate transition structure. To ensure consistency in the division boundaries of the three types of structures, both the first and second dividing lines are centered on the image coordinates of the corresponding reference point and extend to both sides of the boundary with the same sampling interval to the boundary position of the candidate structure.

[0024] After determining the structure on the sun-facing side, the depth of intrusion was determined: In this embodiment, the boundary between the shading intrusion boundary and the light-facing structure is determined as the intrusion initiation position. Subsequently, the edge response values ​​in the light-facing structure are detected point by point along the incident direction of the dominant light source. A preset edge missing threshold is determined based on the statistical results of the edge response value distribution of the light-facing structure and the mean of the edge response values ​​in the neighborhood of the intrusion initiation position. When the edge response value is continuously lower than the preset edge missing threshold, the corresponding sampling position is determined to be in a continuous edge missing state. When the edge response values ​​of multiple subsequent consecutive sampling positions are again continuously higher than the preset edge missing threshold, and there is no inserted sampling position between these multiple consecutive sampling positions where the edge response value is again lower than the preset edge missing threshold, the sampling position farthest along the incident direction of the dominant light source before being higher than the preset edge missing threshold is determined as the termination position of the continuous edge missing state. Simultaneously, the changes in the closure relationship of the structural boundary on the light-facing side are detected point by point along the incident direction of the dominant light source. Specifically, the connection interval between adjacent boundary sampling points on both sides of the current sampling position is first obtained, and a preset closure distance threshold is determined based on the statistical results of the connection interval of the sampling points on the light-facing side of the structural boundary. When the connection interval between adjacent boundary sampling points is continuously less than the preset closure distance threshold, and the edge response value between adjacent boundary sampling points remains higher than the preset edge threshold in step one at multiple consecutive sampling positions, it is determined that the boundary reconnects at that position. The sampling position farthest along the incident direction of the dominant light source before reconnection is determined as the termination position of the continuous closure interruption state. Subsequently, the termination position of the continuous edge missing state and the termination position of the continuous closure interruption state, which is farther from the invasion start position in the incident direction of the dominant light source, is determined as the invasion termination position, and the extension length along the incident direction of the dominant light source between the invasion start position and the invasion termination position is determined as the invasion depth.

[0025] Step 3: The candidate structures of the monitoring targets whose invasion depth reaches the preset conditions are identified as damaged structures for monitoring and interpretation. Based on the invasion depth, the light-facing side structure of the damaged structure is shielded and released to obtain the light-facing side restored structure. Specifically, based on the light-facing structure, the incident direction of the dominant light source, the depth of invasion, the starting position of invasion, and the ending position of invasion obtained in step two, it is determined whether the candidate structure of the monitoring target meets the preset conditions: In this embodiment, the depth of intrusion is first compared with the extension length of the structure on the light-facing side along the incident direction of the dominant light source. Specifically, the starting and ending boundary positions of the structure on the light-facing side are determined along the incident direction of the dominant light source, and the extension length between them is determined as the extension length of the structure on the light-facing side along the incident direction of the dominant light source. Subsequently, the ratio of the depth of intrusion to the extension length is calculated. To determine the preset ratio threshold, the edge response values, changes in closure relationships, and the proportion of intrusion depth at multiple sampling positions distributed along the incident direction of the dominant light source within the candidate structure of the current monitoring target are statistically analyzed to obtain the distribution result of the intrusion ratio of the structure on the light-facing side. Then, based on the distribution result of the intrusion ratio corresponding to multiple sampling positions, the concentrated variation interval of the intrusion ratio is statistically analyzed, and the preset ratio threshold is determined based on the statistical characteristics of the concentrated variation interval. When the ratio of the depth of intrusion to the extension length reaches the preset ratio threshold, the depth of intrusion is determined to meet the depth condition. Subsequently, the termination positions of the continuous edge loss state and the continuous closure interruption state in the structure facing the light were determined, and the continuity of the sampling position sequence between them was analyzed along the incident direction of the dominant light source. When there is no interruption segment between the termination positions of the continuous edge loss state and the termination positions of the continuous closure interruption state where the edge response is continuously enhanced again and the boundary is re-formed into a stable closure relationship, this position interval is determined as the same intruded segment, and the structure facing the light is determined to meet the damage consistency condition. When the intrusion depth meets the depth condition and the structure facing the light meets the damage consistency condition, the candidate structure of the monitoring target is determined as the monitored and interpreted damaged structure.

[0026] After the candidate structure of the monitored target is determined to be a damaged structure for monitoring and interpretation, the shielding release section, as well as the outer release boundary and the inner release boundary, are determined: In this embodiment, the sampling segment between the intrusion start position and the intrusion end position along the incident direction of the dominant light source is defined as the masking release segment. Subsequently, starting from the intrusion end position, the highlight coverage state of each sampling position within the masking release segment is detected point by point along the direction toward the intrusion start position. Specifically, the local brightness value of each sampling position is obtained, and the high brightness coverage state determination rule disclosed in step one is used to determine whether each sampling position is in a high brightness coverage state. When the high brightness coverage state changes from continuous existence to interruption, the interruption position is determined as the first interruption position of the high brightness coverage state, and the sampling position located outside and adjacent to the first interruption position of the high brightness coverage state is determined as the outer release boundary. Next, starting from the initial location of the invasion, the edge visibility status of each sampling location within the masked release segment is detected point by point along the direction towards the termination location of the invasion. Specifically, the edge response value of each sampling location is obtained, and the edge visibility status determination rule disclosed in step one is used to determine whether each sampling location is in an edge visible state. When the edge visibility status changes from continuous absence to continuous visibility, the first continuously visible sampling location is determined as the inner release boundary. Subsequently, the sampling location between the outer release boundary and the inner release boundary is determined as the current release segment.

[0027] Perform phased reduction processing on the current release segment and dynamically update the inner release boundary: In this embodiment, the brightness values ​​corresponding to each sampling position within the current release segment are reduced in stages according to their distance from the initial location of the invasion, with a preset decreasing amplitude. Specifically, the local brightness value distribution of each sampling position within the current release segment is first statistically analyzed to obtain the average brightness, peak brightness, and brightness change gradient of the current release segment. Then, based on the brightness change gradient distribution of adjacent sampling positions within the current release segment, a preset decreasing amplitude is determined, causing the brightness values ​​corresponding to sampling positions closer to the initial location of the invasion to be reduced first, followed by the brightness values ​​corresponding to sampling positions farther from the initial location of the invasion, thereby gradually receding the high-brightness coverage along the incident direction of the dominant light source. After each stage of reduction, the edge response values ​​of each sampling position within the current release segment are reacquired, and the first sampling position where the edge visibility state changes from continuous absence to continuous visibility is redefined as the updated inner release boundary. The preset reduction range is not a fixed constant, but is determined based on the average reduction requirement of the brightness value of each sampling position in the current release segment. Specifically, the difference between the local average brightness value of the sampling position near the outer release boundary and the local average brightness value of the sampling position near the inner release boundary of the current release segment is used as the brightness difference benchmark. Then, the reduction range for each reduction is determined based on the brightness difference benchmark and the length of the current release segment.

[0028] Finally, in this embodiment, the stopping condition is determined based on the updated inner release boundary, and the sunlight-facing recovery structure is output, including: First, the initially determined inner release boundary is taken as the original inner release boundary. Then, the sampling position sequence between the intrusion initiation position and the original inner release boundary along the incident direction of the dominant light source is determined as the back-off judgment section. Subsequently, the edge response value of each sampling position is obtained point by point within the back-off judgment section, and a continuous visible edge sampling position sequence that does not overlap with the shielding release section and is located outside the original inner release boundary along the incident direction of the dominant light source is selected in the light-facing structure as the recovery reference section. The edge response values ​​of each sampling position within the recovery reference section are statistically analyzed to obtain the average edge response value of the recovery reference section, and the average edge response value is determined as the edge response recovery benchmark. Afterward, the edge response values ​​of each sampling position within the back-off judgment section are compared with the edge response recovery benchmark. When the edge response values ​​of multiple consecutive sampling positions along the incident direction of the dominant light source all reach the edge response recovery benchmark, the sampling position closest to the intrusion initiation position among the multiple consecutive sampling positions is determined as the preset back-off position. The number of consecutive sampling positions is determined based on the number of sampling positions between the initial location of the invasion and the original inner release boundary, and their distribution density along the incident direction of the dominant light source. Subsequently, after each step of reduction processing, the relative position of the updated inner release boundary and the preset retreat position is compared; when the updated inner release boundary moves to or crosses the preset retreat position, the step reduction of the brightness values ​​corresponding to each sampling position in the current release segment is stopped, and the processed light-facing side structure is determined as the light-facing side recovery structure.

[0029] Step 4: Extract the residual structural segments of the backlight side structure from the damaged structure in the monitoring and interpretation, determine the structural closure benchmark based on the residual structural segments, and perform closure and restoration processing on the missing structural segments of the intermediate transition structure in the light-facing restored structure and the damaged structure in the monitoring and interpretation based on the structural closure benchmark to obtain the monitoring and interpretation restored structure. Specifically, based on the backlight-side structure already defined in step two and the frontlight-side restored structure already defined in step three, the remaining structural segments are determined: In this embodiment, the edge response values ​​and boundary closure relationships of each sampling position are first acquired point by point on the boundary of the backlight-side structure, and the continuity between adjacent sampling positions is checked along the extension sequence of the backlight-side structure boundary. When the edge response values ​​of adjacent sampling positions are continuously higher than the preset edge threshold determined in step one, and there is no boundary closure interruption position between adjacent sampling positions, the continuous sampling position sequence is determined as a continuous visible edge segment. Then, the continuous visible edge segments with the number of continuous sampling positions reaching the preset retention length condition are determined as residual structure segments. The preset retention length condition is determined based on the statistical results of the number of sampling positions of all continuous visible edge segments on the backlight-side structure boundary. Specifically, the number of sampling positions of each continuous visible edge segment is sorted, and the continuous visible edge segments above the median length are determined as continuous visible edge segments that meet the preset retention length condition.

[0030] Based on the residual structural segments, the corresponding landing points are obtained and the statistical fluctuation range used for sequential trimming is determined: After determining the residual structure segments, the boundary orientation of each sampling position and the sampling interval of each adjacent sampling position along the extension direction of the residual structure segments are determined sequentially along the extension sequence of the residual structure segments. Specifically, for each sampling position, the image coordinates of the previous and next sampling positions are obtained, and the boundary orientation of the current sampling position is determined based on the coordinate changes between the previous, current, and next sampling positions. Then, the sampling interval of each adjacent sampling position along the extension direction of the residual structure segments is determined based on the coordinate differences of each adjacent sampling position along the extension direction of the residual structure segments. Afterward, taking the opposite direction of the incident direction of the dominant light source as the extension direction, the visible edge positions in the intermediate transition structure and the light-facing recovery structure are searched point by point along the extension direction from each sampling position in the residual structure segments. During the point-by-point search, the edge response values ​​of each sampling position on the extension path are obtained sequentially, and the first sampling position that satisfies the edge visibility state determination rule in step one is determined as the first visible edge position, and then the first visible edge position is determined as the corresponding landing point. To establish a judgment boundary for subsequent sequential adjustments, all adjacent sampling position pairs in the residual structural segment are used as a statistical data set. The boundary orientation change and sampling interval change between each adjacent sampling position are statistically analyzed to obtain the boundary orientation change statistical sequence and the sampling interval change statistical sequence. Then, the deviation of each data in the boundary orientation change statistical sequence and the sampling interval change statistical sequence from its median value is statistically analyzed, and the allowable deviation intervals on both sides of the median value are determined based on the deviation distribution. Subsequently, the allowable fluctuation range of the boundary orientation and the allowable fluctuation range of the sampling interval are determined based on the median value and the corresponding allowable deviation interval, respectively. The allowable fluctuation range of the boundary orientation and the allowable fluctuation range of the sampling interval are jointly determined as the statistical fluctuation range.

[0031] Based on the statistical fluctuation range, the sequence of the corresponding landing points is adjusted and the structural closure benchmark is determined: After obtaining the corresponding landing points for each sampling position, the relative positions of each corresponding landing point are sequentially adjusted according to the extension order of the residual structural segments. Specifically, the landing point corresponding to the first sampling position in the residual structural segment is taken as the starting adjustment landing point; then, the directional change and sampling interval change between the subsequent adjacent corresponding landing points and the previous adjustment landing point are obtained sequentially, and these directional change and sampling interval change are compared with the aforementioned statistical fluctuation range. If the directional change between the current corresponding landing point and the previous adjustment landing point exceeds the allowable fluctuation range of the boundary direction, the position of the current corresponding landing point is adjusted along the direction perpendicular to the incident direction of the dominant light source; the adjustment step size is one sampling interval unit of the deviation between the current corresponding landing point and the previous adjustment landing point; if the sampling interval change between the current corresponding landing point and the previous adjustment landing point exceeds the allowable fluctuation range of the sampling interval, the position of the current corresponding landing point is adjusted in the opposite direction of the incident direction of the dominant light source; the adjustment step size is one sampling interval unit of the interval deviation between the current corresponding landing point and the previous adjustment landing point. If the current corresponding landing point has both directional change and sampling interval change that exceed the statistical fluctuation range, first perform position adjustment on the directional change, then perform position adjustment on the sampling interval change, and recalculate whether the two have fallen back to the statistical fluctuation range after each round of adjustment. When both fall back to within the statistical fluctuation range, stop adjusting the current corresponding landing point and determine the adjusted current position as the current trimmed landing point; then continue to perform the same sequential trimming for the next corresponding landing point. After completing the sequential trimming of all corresponding landing points in the above manner, the sequence of sequentially trimmed corresponding landing points is determined as the structural closure benchmark.

[0032] Based on the structural closure benchmark, the missing structural segments in the restored structure on the sun-facing side and the intermediate transition structure are closed and restored to obtain the monitored and interpreted restored structure.

[0033] In this embodiment, the corresponding landing points after modification in the structural closure reference are used as closure restoration control points. The system sequentially checks whether there are missing structural segments between the restored structural boundary on the light-facing side and the intermediate transition structural boundary. When there is a boundary closure interruption position between adjacent closure restoration control points, the interrupted segment is identified as a missing structural segment, and the missing structural segment is restored segment by segment along the extension order of the structural closure reference. Specifically, starting from the closure restoration control point near the backlight side of the structure, the restoration boundary sampling position corresponding to the current missing structural segment is generated point by point according to the order between adjacent closure restoration control points, and the restoration boundary sampling position is sequentially moved closer to the landing point position corresponding to the structural closure reference. After the restoration boundary sampling position of the current missing structural segment is generated, adjacent restoration boundary sampling positions are connected according to the boundary extension order to form a restoration boundary segment. The same process is then performed on the next missing structural segment until all missing structural segments between the restored structure on the light-facing side and the intermediate transition structure are restored. When the restored boundary remains connected at multiple consecutive sampling positions, and the edge response value between adjacent sampling positions after restoration is continuously higher than the preset edge threshold in step one, the restored light-facing side restoration structure, the intermediate transition structure, and the backlight side structure that maintains a continuous connection with it are jointly determined as the monitoring and interpretation restoration structure.

[0034] Step 5: Perform a closure determination on the monitored and restored structure. If the closure determination result does not meet the preset closure conditions, adjust the occlusion release process in Step 3 and the closure restoration process in Step 4. If the closure determination result meets the preset closure conditions, output the enhanced image. Specifically, based on the monitoring and interpretation recovery structure obtained in step four, the structural closure detection object and preset closure conditions for closure determination are determined: In this embodiment, a continuous boundary sampling position sequence formed by the backlight side structure, the light-facing side recovery structure, and the intermediate transition structure is first extracted from the monitoring and interpretation recovery structure, and this continuous boundary sampling position sequence is determined as the structure closure detection object. Subsequently, along the extension order of the structure closure detection object, the edge response value, the connection interval between adjacent sampling positions, and the directional change between adjacent sampling positions are obtained point by point. Then, according to the structure closure benchmark determined in step four, the benchmark edge response value, benchmark connection interval, and benchmark directional change at the corresponding positions of the structure closure benchmark are obtained respectively. After that, the edge response value of the structure closure detection object is compared with the benchmark edge response value, the connection interval between adjacent sampling positions is compared with the benchmark connection interval, and the directional change between adjacent sampling positions is compared with the benchmark directional change. When the edge response value continuously reaches the benchmark edge response value requirement, the connection interval continuously falls within the allowable range of the benchmark connection interval, and the directional change continuously falls within the allowable range of the benchmark directional change, it is determined that the monitoring and interpretation recovery structure meets the closure requirement of the corresponding position. To determine the preset closure conditions, the distribution of connection intervals and orientation changes of all adjacent reference positions in the structural closure reference are statistically analyzed to obtain closure determination statistics. The minimum number of consecutive samples that continuously satisfy connection stability and orientation stability in the closure determination statistics is used as the continuous determination number requirement in the preset closure conditions. Through this step, subsequent closure determination can be based on the corresponding comparison between the restored structure and the structural closure reference, rather than relying solely on a single edge strength or a single connection length.

[0035] Based on preset closure conditions, a closure determination is performed on the monitoring and interpretation recovery structure: In this embodiment, the sampling positions are sequentially checked to ensure they meet the closure requirements along the extension sequence of the structural closure detection object. When multiple consecutive sampling positions meet the corresponding requirements for edge response value, connection interval, and direction change, and the number of these multiple consecutive sampling positions reaches the consecutive judgment quantity requirement in the preset closure conditions, the sequence of consecutive sampling positions is determined as a valid closure segment. Then, the distribution of all valid closure segments in the structural closure detection object is statistically analyzed. When a valid closure segment covers all closure segments defined by the adjacent adjusted corresponding landing points in the structural closure benchmark within the monitored and restored structure, and there are no boundary closure interruption positions where the connection interval exceeds the allowable range again between valid closure segments, the closure judgment result is determined to meet the preset closure conditions. When a valid closure segment does not cover all closure segments, or there are boundary closure interruption positions where the connection interval exceeds the allowable range again between valid closure segments, the closure judgment result is determined to not meet the preset closure conditions. Through this step, the "closure judgment result" can be implemented as a comprehensive judgment result on the continuity, connectivity, and direction consistency of the restored structure, thus maintaining consistency with the determination method of the "structural closure benchmark" in step four.

[0036] If the closure determination result does not meet the preset closure conditions, the occlusion release process in step three and the closure restoration process in step four will be adjusted accordingly: In this embodiment, when the closure determination result does not meet the preset closure conditions, the abnormal segment in the monitoring and judgment restoration structure that does not meet the closure requirements is first located. Specifically, along the extension sequence of the structure closure detection object, the sampling position sequence where the edge response value does not reach the reference edge response value requirement, the connection interval exceeds the allowable range of the reference connection interval, or the direction change exceeds the allowable range of the reference direction change is found, and the sampling position sequence is determined as the abnormal segment. Subsequently, based on the relative positional relationship between the abnormal segment and the light-facing side restoration structure, it is determined whether the shielding release process in step three needs to be adjusted. When the abnormal segment is close to the light-facing side restoration structure and the edge response value in the abnormal segment is continuously lower than the edge response restoration reference determined in step three, the brightness value reduction range or reduction magnitude corresponding to the current release segment in step three is extended to the insufficiently released part between the invasion start position and the original inner release boundary, and the updated inner release boundary is re-acquired. Simultaneously, based on the relative position of the abnormal section and the intermediate transition structure, it is determined whether the closure recovery process in step four needs adjustment. If the abnormal section is located between the intermediate transition structure and the light-facing recovery structure, and there are still boundary closure interruption positions between adjacent closure recovery control points, then the abnormal section is re-included in the missing structure segment processing range of step four. After adjusting the recovery order between the closure recovery control points according to the original structural closure benchmark, the segment-by-segment closure recovery process is re-executed. After completing the above adjustments, the monitoring and interpretation recovery structure is regenerated, and the closure judgment is executed again.

[0037] When the closure determination result meets the preset closure condition, an enhanced image is output: In this embodiment, when the closure determination result meets the preset closure condition, the boundary sampling position, edge response value and brightness value corresponding to the monitoring judgment recovery structure are written into the corresponding sampling position area in the input monitoring image to obtain an enhanced image and output it.

[0038] In one embodiment, after the monitoring and interpretation recovery structure is written into the corresponding area of ​​the input monitoring image, the boundary between the recovery area and the original area can be further processed according to the brightness changes and edge response changes on both sides of the boundary of the recovery area.

[0039] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for enhancing surveillance images in low-light environments, characterized in that, include: Step 1: Acquire the input monitoring image, determine the dominant light source area, and extract the highlight coverage distribution information and edge visible distribution information outward from the dominant light source area. Determine the occlusion intrusion boundary based on the continuous coverage termination position in the highlight coverage distribution information and the continuous visible termination position in the edge visible distribution information. Step 2: Extract the candidate structures of the monitoring targets that intersect with or are located inside the shielding intrusion boundary. Divide the candidate structures of the monitoring targets into the light-facing side structure, the back-light side structure, and the intermediate transition structure according to the incident direction of the dominant light source. Determine the intrusion depth based on the intersection position of the shielding intrusion boundary and the light-facing side structure, as well as the termination position of the continuous edge missing state or the termination position of the continuous closed interruption state in the light-facing side structure. Step 3: The candidate structures of the monitoring targets whose invasion depth reaches the preset conditions are identified as damaged structures for monitoring and interpretation. Based on the invasion depth, the light-facing side structure of the damaged structure is shielded and released to obtain the light-facing side restored structure. Step 4: Extract the residual structural segments of the backlight side structure from the damaged structure in the monitoring and interpretation, determine the structural closure benchmark based on the residual structural segments, and perform closure and restoration processing on the missing structural segments of the intermediate transition structure in the light-facing restored structure and the damaged structure in the monitoring and interpretation based on the structural closure benchmark to obtain the monitoring and interpretation restored structure. Step 5: Perform a closure judgment on the monitoring and interpretation recovery structure. If the closure judgment result does not meet the preset closure conditions, adjust the masking release process in Step 3 and the closure recovery process in Step 4. When the closure determination result meets the preset closure conditions, an enhanced image is output.

2. The method for enhancing surveillance images in low-light environments according to claim 1, characterized in that, The method for determining the occlusion intrusion boundary in step one includes: The bright connected regions in the input monitoring image are extracted, and the bright connected regions with the largest brightness peak and continuous outward bright coverage along multiple preset directions are identified as the dominant light source regions. Starting from the sampling position where the brightness peak is located in the dominant light source region, the high brightness coverage state and edge visibility state of each sampling position are extracted sequentially outward along multiple preset directions to generate high brightness coverage distribution information and edge visibility distribution information corresponding to each preset direction. In each preset direction, the outermost sampling position that is continuously in a highlighted coverage state is determined as the continuous coverage termination position, and the innermost sampling position that is continuously in an edge-visible state is determined as the continuous visibility termination position. Within the segment between the continuous coverage termination position and the continuous visibility termination position, the first sampling position where the edge visibility state begins to change from continuous visibility to continuous weakening while the corresponding highlight coverage state remains continuous is determined as the occlusion intrusion position in the preset direction. When the difference in radial distance between the shielding intrusion positions in adjacent preset directions is less than a preset distance threshold, it is determined that the shielding intrusion positions in adjacent preset directions meet the continuous distribution condition. By sequentially connecting the shielding intrusion positions in adjacent preset directions that meet the continuous distribution condition, the shielding intrusion boundary is obtained.

3. The method for enhancing surveillance images in low-light environments according to claim 1, characterized in that, The method for determining the incident direction of the dominant light source in step two includes the following steps: By performing a localization analysis on the brightness distribution within the dominant light source region, the location of the brightness peak within the dominant light source region is obtained; By analyzing the location of the boundary between the candidate structure of the monitoring target and the boundary of the intrusion, the boundary position closest to the brightness peak position is obtained as the contact position on the light-facing side. Starting from the contact position on the light-facing side, continuously missing edge response segments are detected point by point into the candidate structure of the monitored target along the direction from the peak brightness position to the contact position on the light-facing side. The termination position of the continuously missing edge response segment is obtained as the intrusion extension position. On the boundary of the candidate structure of the monitoring target, starting from the affected extension position, search for continuous visible edge segments along the side away from the brightness peak position, and determine the starting position of the first continuous visible edge segment obtained by the search as the backlight side reserved position; A first directional basic quantity is formed from the brightness peak position to the contact position on the light-facing side, and a second directional basic quantity is formed from the contact position on the light-facing side to the intruded extension position; When the directional deviation between the first directional base quantity and the second directional base quantity is less than a preset directional deviation threshold, the second directional base quantity is used as the dominant light source incident direction. When the directional deviation between the first directional base quantity and the second directional base quantity is greater than or equal to the preset directional deviation threshold, the midpoint of the line connecting the affected extension position and the backlight side retention position is taken as the correction position, and the direction from the brightness peak position to the correction position is taken as the incident direction of the dominant light source.

4. The method for enhancing surveillance images in low-light environments according to claim 3, characterized in that, The method for dividing the candidate structure of monitoring targets in step two includes: The direction of the dominant light source incident is used as the structural dividing direction; the first dividing line is formed along the direction perpendicular to the dominant light source incident direction with the affected extension position as the reference point; and the second dividing line is formed along the direction perpendicular to the dominant light source incident direction with the backlight side retention position as the reference point. The portion of the candidate structure of the monitoring target located on the side facing the dominant light source area of ​​the first dividing line is determined as the light-facing side structure; the portion of the candidate structure of the monitoring target located on the side away from the dominant light source area of ​​the second dividing line is determined as the backlight side structure; and the portion of the candidate structure of the monitoring target located between the first dividing line and the second dividing line is determined as the intermediate transition structure.

5. The method for enhancing surveillance images in low-light environments according to claim 4, characterized in that, The method for determining the depth of invasion in step two includes: The boundary between the shielding intrusion boundary and the sun-facing structure is determined as the initiation point of the intrusion. Determine the termination position of the continuous edge missing state in the structure on the sun-facing side; Determine the termination position of the continuous closed interruption state in the structure facing the light; The position that is farther from the start of the invasion in the direction of the dominant light source incident is determined as the invasion termination position, between the termination position of the continuous edge missing state and the termination position of the continuous closed interruption state. The depth of invasion is defined as the length of the extension along the incident direction of the dominant light source between the intrusion start position and the intrusion termination position.

6. The method for enhancing surveillance images in low-light environments according to claim 1, characterized in that, The preset conditions in step three are: The depth of the intrusion was compared with the length of the structure on the light-facing side along the incident direction of the dominant light source. When the ratio of the penetration depth to the extension length reaches a preset ratio threshold, the penetration depth is determined to meet the depth condition. Determine the termination positions of continuous edge missing state and continuous closed interruption state in the structure facing the light; When the termination position of the continuous edge missing state and the termination position of the continuous closed interruption state are located in the same affected section, it is determined that the structure on the sun-facing side meets the damage consistency condition. When the depth of the intrusion meets the depth condition and the structure on the sun-facing side meets the damage consistency condition, the candidate structure of the monitoring target is determined as the damaged structure for monitoring and interpretation.

7. The method for enhancing surveillance images in low-light environments according to claim 1, characterized in that, Step 3, the masking release process, includes the following steps: The sampling section along the incident direction of the dominant light source between the intrusion start position and the intrusion termination position is defined as the shielding release section. Starting from the termination point of the invasion, the highlight coverage status of each sampling position in the masking release section is detected point by point along the direction towards the start point of the invasion. The first sampling position where the highlight coverage status changes from continuous to interrupted is determined, and the sampling position located outside the first sampling position and adjacent to it is determined as the outer release boundary. Starting from the initial location of the invasion, the edge visibility status of each sampling location within the occluded release section is detected point by point along the direction toward the termination location of the invasion. The first sampling location where the edge visibility status changes from continuous absence to continuous visibility is determined as the inner release boundary. The sampling location between the outer release boundary and the inner release boundary is determined as the current release segment; The brightness values ​​corresponding to each sampling position within the current release segment are reduced in stages according to the order from the nearest to the starting position of the invasion, with a preset decreasing amplitude. After each stage of reduction, the first sampling position where the edge visibility state changes from continuous missing to continuous visible is re-determined as the updated inner release boundary. When the updated inner release boundary moves to the preset back-off position between the invasion start position and the original inner release boundary, the phased reduction of the brightness value corresponding to each sampling position in the current release segment is stopped, and the processed light-facing side structure is determined as the light-facing side recovery structure.

8. The method for enhancing surveillance images in low-light environments according to claim 1, characterized in that, The method for determining the structural closure datum in step four includes: Determine a continuously visible residual structural segment on the boundary of the backlight-side structure; Following the extension sequence of the residual structural segments, the boundary orientation of each sampling location and the sampling interval of each adjacent sampling location along the extension direction of the residual structural segments are determined sequentially. Using the opposite direction of the incident direction of the dominant light source as the extension direction, the first visible edge position obtained by searching along the extension direction in the intermediate transition structure and the light-facing recovery structure from each sampling position is determined as the corresponding landing point. Based on the boundary orientation change and sampling interval relationship between adjacent sampling positions in the residual structural segment, the relative positions of each corresponding landing point are sequentially adjusted so that the adjusted adjacent corresponding landing points maintain the same orientation change and sampling interval relationship as the residual structural segment. The sequence of corresponding landing points after sequential adjustment is determined as the structural closure benchmark.

Images