Obstruction evolution spectrum-based field of view failure discrimination method and system

Abstract

The application discloses a field-of-view failure discrimination method and system based on an occlusion evolution spectrum, and belongs to the technical field of computer vision monitoring, and comprises the following steps: removing the direct current component from the frame mean of the occlusion occupation amount of each spatial partition in a field of view, to obtain a partition occlusion fluctuation sequence; performing time-frequency joint decomposition on the partition occlusion fluctuation sequence to obtain a time-frequency spectrum, extracting the distribution of the low-frequency component energy of each spatial partition with frame number from the time-frequency spectrum, and constructing an occlusion evolution spectrum; obtaining a cross-partition time sequence correlation matrix group, merging the spatial partitions according to the time sequence correlation magnitude of each partition pair in the cross-partition time sequence correlation matrix group and the spatial adjacent relationship, and obtaining an occlusion spreading area topology diagram; and outputting a field-of-view failure discrimination result. The application unifies the spatial coverage range, energy intensity and spreading dynamic characteristics of the occlusion into the failure discrimination basis, and solves the slow spreading type field-of-view failure missed detection problem caused by the neglect of the occlusion spreading process in the prior art.

Description

Technical Field

[0001] This invention relates to the field of computer vision monitoring technology, specifically to a method and system for determining field of view failure based on occlusion evolution spectrum. Background Technology

[0002] In video surveillance systems, obstruction of the camera's field of view is a common cause of field-of-view failure. Existing methods for determining field-of-view failure typically rely on static assessments of the occupancy status based on pixel statistical features of a single frame or threshold comparisons of global occupancy.

[0003] However, such methods cannot effectively distinguish between the dynamic spread of occlusion in the field of view and brief random occlusion disturbances, resulting in a large number of misjudgments in real-world scenarios.

[0004] For example, when a moving target briefly obstructs a local area within the field of view, methods based on a global occupancy threshold may misclassify it as a field of view failure. Conversely, when the occupancy slowly spreads from the edge of the field of view towards the center, eventually leading to complete field of view failure, the method may fail to trigger a failure alarm for an extended period due to the small increment of occupancy at each moment, resulting in missed detections. Existing methods lack the ability to analyze the propagation patterns of occupancy between spatial partitions and cannot extract effective failure criteria from the temporal evolution characteristics of occupancy. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention provides a method and system for determining field failure based on occlusion evolution spectrum.

[0006] Therefore, the technical problem solved by the present invention is that the existing field failure discrimination method cannot effectively characterize the dynamic spread process of occlusion between different spatial partitions of the field of view, resulting in missed detection of slow-spreading occlusion failure and misjudgment of random and transient occlusion disturbances.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a field-of-view failure discrimination method based on occlusion evolution spectrum, comprising the following steps: removing the DC component of the occlusion occupancy of each spatial partition within the field of view according to the frame mean to obtain a partition occlusion fluctuation sequence; performing time-frequency joint decomposition on the partition occlusion fluctuation sequence to obtain a time spectrum, extracting the distribution of low-frequency component energy of each spatial partition with frame from the time spectrum to construct an occlusion evolution spectrum; performing cross-partition time-series correlation calculation on the low-frequency energy time series of each spatial partition in the occlusion evolution spectrum to obtain a cross-partition time-series correlation matrix group, merging the spatial partitions according to the time-series correlation magnitude and spatial adjacency relationship of each partition pair in the cross-partition time-series correlation matrix group to obtain an occlusion spread area topology map; normalizing the spatial coverage of the occlusion spread area in the occlusion spread area topology map and the low-frequency energy proportion of the corresponding spatial partition in the occlusion evolution spectrum, respectively, and judging the field-of-view failure state based on the comprehensive magnitude of the normalized spatial coverage and low-frequency energy proportion, and outputting the field-of-view failure discrimination result.

[0008] As a preferred embodiment of the field-of-view failure discrimination method based on occlusion evolution spectrum described in this invention, the step of performing the time-frequency joint decomposition includes: segmenting the partitioned occlusion fluctuation sequence according to a preset frame length; applying a short time window function to the occlusion occupancy time sequence within each segment; performing frequency decomposition on the occupancy time sequence within each short time window to obtain the full-band frequency component energy distribution of each spatial partition at each frame position; extracting low-frequency features from the full-band frequency component energy distribution to obtain the low-frequency feature sequence of each spatial partition; and constructing an occlusion evolution spectrum from the structured record with the spatial partition as the row index and the low-frequency feature sequence as the row content.

[0009] As a preferred embodiment of the field-of-view failure discrimination method based on occlusion evolution spectrum described in this invention, the step of extracting the low-frequency features includes: arranging the energy components below the occlusion dynamic change cutoff frequency in the full-band frequency component energy distribution by frame to obtain the time sequence of low-frequency component energy in each spatial partition, and storing it as a low-frequency feature sequence in the occlusion evolution spectrum; normalizing the full-band frequency component energy distribution in each short time window according to the total energy of the full-band, to obtain the normalized frequency component energy distribution in each short time window; calculating the centroid position of each frequency component in the low-frequency range of the normalized frequency component energy distribution by frequency weighting, to obtain the low-frequency centroid position sequence of each frame in each spatial partition; performing difference on the low-frequency centroid position sequence between adjacent frames to obtain the low-frequency centroid drift sequence of each spatial partition, and storing the low-frequency centroid drift sequence together in the occlusion evolution spectrum.

[0010] As a preferred embodiment of the field failure discrimination method based on occlusion evolution spectrum described in this invention, the step of performing the cross-segment temporal correlation calculation includes: measuring the cross-segment temporal correlation of the low-frequency component energy time series of any two spatial partitions in the occlusion evolution spectrum within the allowable time lag range, and obtaining the temporal correlation magnitude and time lag of each partition pair; constructing a temporal correlation magnitude matrix with the temporal correlation magnitude of each partition pair, and constructing a time lag matrix with the time lag of each partition pair; and recording the temporal correlation magnitude matrix and the time lag matrix together as a cross-segment temporal correlation matrix group.

[0011] As a preferred embodiment of the field-of-view failure discrimination method based on occlusion evolution spectrum described in this invention, the step of performing the cross-segment temporal correlation measurement includes: calculating a normalized cross-correlation function for the low-frequency component energy time series of any two spatial partitions in the occlusion evolution spectrum within the allowable time lag range, obtaining the normalized cross-correlation function curve of each partition based on the low-frequency component energy time series, and extracting the first correlation coefficient and the first time lag at the peak value; calculating a normalized cross-correlation function for the low-frequency centroid drift sequence of any two spatial partitions in the occlusion evolution spectrum within the allowable time lag range, obtaining the normalized cross-correlation function curve of each partition based on the low-frequency centroid drift sequence, and extracting the second correlation coefficient and the second time lag at the peak value; summing the low-frequency component energy time series of each spatial partition on the time axis to obtain the low-frequency component energy time series of each spatial partition. The total energy of the components is determined by the proportion of the total low-frequency component energy in the total low-frequency component energy of all spatial partitions as the energy dominance. The first correlation coefficient based on the low-frequency component energy time series and the second correlation coefficient based on the low-frequency centroid drift sequence are weighted and summed according to the mean energy dominance of the corresponding spatial partitions at both ends to obtain the comprehensive time series correlation magnitude of each partition pair. The first and second time delay quantities are checked for sign consistency: if they have the same sign, they are weighted and summed according to the mean energy dominance of the corresponding spatial partitions at both ends to obtain the time delay quantity of each partition pair; if they have opposite signs, the sign of the time delay quantity corresponding to the one with the higher mean energy dominance of the corresponding spatial partitions at both ends is used as the benchmark, and the sign of the other time delay quantity is aligned before being weighted and summed according to the mean energy dominance to obtain the time delay quantity of each partition pair.

[0012] As a preferred embodiment of the field-of-view failure discrimination method based on occlusion evolution spectrum described in this invention, the step of merging the spatial partitions includes: constructing a spatial adjacency graph with spatial partitions as nodes and spatial adjacency relationships as hard connection constraints; extracting the corresponding edge weights from the temporal correlation magnitude matrix of the cross-partition temporal correlation matrix for each adjacent partition pair in the spatial adjacency graph to obtain a spatial adjacency weighted graph; merging the connecting edges in the spatial adjacency weighted graph stepwise according to the edge weight magnitude from high to low, merging the spatial partitions in the same connected domain after stepwise merging into an occlusion propagation region; averaging the time delay values ​​of partition pairs within each occlusion propagation region according to the temporal correlation magnitude to obtain the representative time delay value within each occlusion propagation region; averaging the time delay values ​​of partition pairs crossing the boundaries of different occlusion propagation regions according to the temporal correlation magnitude to obtain the representative time delay value between each occlusion propagation region; and performing a first encoding on the propagation temporal relationship between all occlusion propagation regions based on the representative time delay value between the regions to obtain an occlusion propagation region topology graph.

[0013] As a preferred embodiment of the field failure discrimination method based on occlusion evolution spectrum described in this invention, the step of performing the first encoding includes: assigning propagation direction attributes to the connecting edges between each occlusion spread region according to the positive and negative directions of the time delay represented by the interval, thereby obtaining a directed topology graph; taking the longest directed path in the directed topology graph that starts from each occlusion spread region and reaches along the propagation direction, taking the sum of the total number of spatial partitions covered by all occlusion spread regions on the path as the path spatial span; taking the ratio of the path spatial span to the cumulative value of the time delay represented by the interval of each pair of adjacent occlusion spread regions on the longest directed path as the spread rate of each occlusion spread region; normalizing the spread rate according to the maximum value of the spread rate of all occlusion spread regions to obtain a normalized spread rate, and recording the normalized spread rate in each corresponding occlusion spread region node in the occlusion spread region topology graph.

[0014] As a preferred embodiment of the field-of-view failure discrimination method based on occlusion evolution spectrum described in this invention, the step of discerning the field-of-view failure state includes: summing the low-frequency component energy time series of each spatial partition in the occlusion evolution spectrum on the time axis to obtain the total low-frequency component energy of each spatial partition; normalizing the total low-frequency component energy according to the total energy of the occlusion fluctuation sequence of the partition on the time axis to obtain the low-frequency energy ratio of each spatial partition; normalizing the low-frequency energy ratio according to the maximum value of the low-frequency energy ratio of all spatial partitions to obtain the normalized low-frequency energy of each spatial partition; and aggregating the normalized low-frequency energy of all spatial partitions in each occlusion spread region topology map by the number of spatial partitions to obtain the low-frequency energy of each occlusion spread region. The following parameters are used to determine the spatial coverage of each occlusion spread area in the topology map: The spatial coverage normalization is obtained by normalizing the spatial coverage range of each occlusion spread area according to the total number of field-of-view partitions. The mean spatial center coordinates of all spatial partitions within each occlusion spread area are used as the regional center coordinates of that occlusion spread area. The first distance between the regional center coordinates and the field-of-view center coordinates is calculated. The maximum value of the first distance among all occlusion spread areas is used to normalize the first distance of each occlusion spread area. The complement of the normalized first distance is used as the positional importance of each occlusion spread area. The spatial coverage normalization, the low-frequency energy representative quantity, and the normalized spread rate are weighted and accumulated according to the positional importance to obtain the comprehensive magnitude of field-of-view failure. The field-of-view failure judgment result is output based on the comprehensive magnitude of field-of-view failure.

[0015] This invention provides a field failure discrimination system based on occlusion evolution spectrum.

[0016] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a field-of-view failure discrimination system based on occlusion evolution spectrum, comprising:

[0017] The fluctuation sequence extraction unit is used to remove the DC component of the occlusion occupancy of each spatial partition within the field of view based on the frame mean, so as to obtain the partition occupancy fluctuation sequence.

[0018] The occlusion evolution spectrum construction unit is used to perform time-frequency joint decomposition on the partitioned occlusion fluctuation sequence to obtain the time spectrum, extract the distribution of low-frequency component energy of each spatial partition with the frame from the time spectrum, and construct the occlusion evolution spectrum.

[0019] The shading spread region topology map generation unit is used to perform cross-partition temporal correlation calculation on the low-frequency energy time series of each spatial partition in the shading evolution spectrum to obtain a cross-partition temporal correlation matrix group. Based on the temporal correlation magnitude and spatial adjacency relationship of each partition pair in the cross-partition temporal correlation matrix group, the spatial partitions are merged to obtain the shading spread region topology map.

[0020] The field of view failure discrimination unit is used to normalize the spatial coverage of the occlusion spread area in the topology map of the occlusion spread area and the low-frequency energy ratio of the corresponding spatial partition in the occlusion evolution spectrum, respectively, and to judge the field of view failure state based on the comprehensive magnitude of the normalized spatial coverage area and the low-frequency energy ratio, and output the field of view failure discrimination result.

[0021] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the described method for determining field failure based on occlusion evolution spectrum.

[0022] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the field failure discrimination method based on occlusion evolution spectrum.

[0023] The beneficial effects of this invention are as follows: By removing the DC component from the occupancy of each spatial partition and then performing time-frequency joint decomposition, the low-frequency dynamic components of the occupancy are separated from the full-band frequency component energy distribution. An occupancy evolution spectrum is constructed using spatial partitions as row indices, thereby structurally expressing the temporal evolution characteristics of the occupancy and effectively avoiding interference from transient random disturbances in failure detection. Furthermore, by calculating the cross-partition temporal correlation of the low-frequency energy time series of each spatial partition in the occupancy evolution spectrum and merging the partitions based on spatial adjacency relationships, a topological map of the occupancy propagation area is obtained, achieving a structured description of the spatial propagation law of the occupancy.

[0024] Furthermore, based on the normalized spatial coverage, low-frequency energy representativeness, and normalized spread rate of the occlusion spread area, the data is weighted and accumulated according to the location importance of each occlusion spread area to obtain the comprehensive magnitude of the field of view failure. This integrates the spatial coverage range, energy intensity, and spread dynamic characteristics of the occlusion into the failure discrimination criteria, solving the problem of missed detection of slow-spreading field of view failures caused by neglecting the occlusion spread process in existing methods. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 The above is a flowchart of a field failure discrimination method based on occlusion evolution spectrum provided in one embodiment of the present invention. Detailed Implementation

[0027] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0028] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a field-of-view failure discrimination method based on occlusion evolution spectrum, including the following steps:

[0029] The occlusion occupancy of each spatial partition within the field of view is removed by removing the DC component based on the frame mean, resulting in a partition occlusion fluctuation sequence.

[0030] The time-frequency joint decomposition of the partitioned occlusion fluctuation sequence is used to obtain the time spectrum. The distribution of low-frequency component energy of each spatial partition with frame is extracted from the time spectrum to construct the occlusion evolution spectrum.

[0031] Cross-regional temporal correlation calculation is performed on the low-frequency energy time series of each spatial region in the occlusion evolution spectrum to obtain a cross-regional temporal correlation matrix group. Based on the temporal correlation magnitude and spatial adjacency relationship of each pair of regions in the cross-regional temporal correlation matrix group, the spatial regions are merged to obtain the topology map of the occlusion spread region.

[0032] The spatial coverage area of ​​the occlusion spread area in the topology map and the low-frequency energy ratio of the corresponding spatial partition in the occlusion evolution spectrum are normalized respectively. The field of view failure status is judged based on the comprehensive magnitude of the normalized spatial coverage area and the low-frequency energy ratio, and the field of view failure judgment result is output.

[0033] Existing methods for identifying field failures cannot effectively characterize the dynamic spread of occlusion between different spatial partitions of the field of view, leading to missed detections of slow-spreading occlusion failures and misjudgments of random, transient occlusion disturbances.

[0034] By removing the DC component from the occupancy of each spatial partition and then performing time-frequency joint decomposition, the low-frequency dynamic components of the occupancy are separated from the full-band frequency component energy distribution. An occupancy evolution spectrum is constructed using spatial partitions as row indices, thus structurally representing the temporal evolution characteristics of the occupancy and effectively avoiding interference from transient random disturbances in failure detection. Based on this, cross-partition temporal correlation calculations are performed on the low-frequency energy time series of each spatial partition in the occupancy evolution spectrum. By merging the partitions based on spatial adjacency relationships, a topological map of the occupancy spread area is obtained, achieving a structured description of the spatial spread pattern of the occupancy.

[0035] Furthermore, based on the normalized spatial coverage, low-frequency energy representativeness, and normalized spread rate of the occlusion spread area, the data is weighted and accumulated according to the location importance of each occlusion spread area to obtain the comprehensive magnitude of the field of view failure. This integrates the spatial coverage range, energy intensity, and spread dynamic characteristics of the occlusion into the failure discrimination criteria, solving the problem of missed detection of slow-spreading field of view failures caused by neglecting the occlusion spread process in existing methods.

[0036] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the previous embodiment, a field-of-view failure discrimination method based on occlusion evolution spectrum is provided, including the following steps:

[0037] S1. The occlusion occupancy of each spatial partition within the field of view is removed by removing the DC component based on the frame mean, resulting in a partition occlusion fluctuation sequence.

[0038] In this step, the field of view covered by the camera is first spatially partitioned. Specifically, the field of view image is uniformly divided into M×N non-overlapping rectangular sub-regions along the row and column directions. Each sub-region is called a spatial partition and is identified by coordinate pairs (i,j), where i∈{1,2,…,M} represents the row number and j∈{1,2,…,N} represents the column number. The granularity of the spatial partitioning is determined based on the field of view resolution and target detection accuracy requirements. The area of ​​each spatial partition should be sufficient to accommodate the pixel range covered by a typical occluded target in a single frame image.

[0039] For a continuously acquired image sequence within the field of view, the occlusion occupancy O(i,j,t) of each spatial partition (i,j) is calculated frame by frame according to frame number t (t=1,2,…,T, where T is the total number of frames). The occupancy occupancy is defined as the ratio of the number of pixels in spatial partition (i,j) that are determined to be occluded to the total number of pixels in that partition in the t-th frame image, and the value range is [0,1].

[0040] The DC component refers to a constant background component in the time series of occupancy that does not change or changes extremely slowly over time, corresponding to zero-frequency or very low-frequency components in the frequency domain. The presence of the DC component will cause interference from the mean bias of the series in the time-frequency joint decomposition results, masking the true dynamic fluctuation characteristics of occupancy in the time series. Therefore, it needs to be removed before time-frequency analysis.

[0041] Time series of occlusion occupancy for each spatial partition (i,j) Calculate its frame mean as follows: :

[0042] ;

[0043] in, This is the time average of the occupancy of the spatial partition during the entire observation period, representing the steady-state background level of the partition's occupancy state, denoted by DC component, and T is the total number of frames.

[0044] Subsequently, the occupancy amount of each spatial partition (i,j) at each frame t is subtracted from its frame mean to obtain the partition occupancy fluctuation sequence after removing the DC component.

[0045] The mean of the partition occlusion fluctuation sequence is zero. A positive value indicates that the occlusion level of the partition in that frame is higher than the long-term average level, while a negative value indicates that it is lower than the long-term average level. This effectively highlights the dynamic change of occlusion occupancy over time and filters out the steady-state bias introduced by long-term static occlusion (such as fixed occlusion at the edge of the camera lens).

[0046] S2. Perform time-frequency joint decomposition on the partitioned occlusion fluctuation sequence to obtain the time spectrum, extract the distribution of low-frequency component energy of each spatial partition with frame from the time spectrum, and construct the occlusion evolution spectrum.

[0047] Specifically, the steps for performing the time-frequency joint decomposition include:

[0048] S2.1. The partition occupancy fluctuation sequence is segmented according to a preset frame length. A short time window function is applied to the occupancy occupancy time sequence in each segment. Frequency decomposition is performed on the occupancy occupancy time sequence in each short time window to obtain the full-band frequency component energy distribution of each spatial partition at each frame position.

[0049] The occlusion fluctuation sequence of each spatial partition (i,j) is segmented according to a preset frame length L. The preset frame length L is the temporal resolution parameter for time-frequency analysis, and its value needs to achieve a balance between temporal resolution and frequency resolution: if L is too small, the frequency resolution is insufficient, and it is impossible to effectively distinguish between low-frequency and high-frequency components; if L is too large, the temporal resolution decreases, and it is impossible to capture the local temporal changes in the occlusion evolution. In practical applications, L can be set according to the video frame rate and the typical duration of the occlusion event. For example, when the video frame rate is 25 frames / second and the shortest duration of the occlusion event is about 2 seconds, L can be set to 32 or 64 frames.

[0050] The segmentation method employs a sliding window strategy, with overlapping frame steps S (S≤L) between adjacent segments to ensure that each frame can obtain the corresponding time-frequency analysis results. For the k-th short time window in the sequence, its starting frame is (k−1)·S+1, and its ending frame is (k−1)·S+L. The corresponding center frame is used as the frame position marker for this short time window.

[0051] A short-time window function is applied to the occlusion wave subsequence within each short-time window to suppress spectral leakage at the segment boundaries. The short-time window function is expressed as:

[0052] ;

[0053] The occlusion fluctuation subsequence within the k-th short-time window after windowing is:

[0054] ;

[0055] For windowed subsequences Calculations were performed to obtain the complex spectral coefficients of each frequency component:

[0056] ;

[0057] Where m is the frequency component index, and the corresponding actual frequency is... , The frame rate for video capture. The energy of each frequency component is: ;

[0058] Because the computation of real-valued sequences has conjugate symmetry, take common Each independent frequency component constitutes the full-band frequency component energy distribution of the k-th short time window at the frame position.

[0059] Repeat the above process for all short time windows to obtain the full-band frequency component energy distribution of each spatial partition (i,j) at each frame position.

[0060] For example, the field of view is divided into 16 spatial partitions of 4×4, the observation period is T=100 frames, and the video frame rate is... Frames per second.

[0061] Assuming a preset frame length L = 32 frames and a frame step size S = 8 frames, the center frames of each short time window are as follows: There are approximately 11 short time windows covering a 100-frame sequence.

[0062] Taking the first short window of spatial partition (2,3) as an example:

[0063] After applying a Hanning window to the occlusion fluctuation values ​​of the 32 frames within the window, a 32-point calculation was performed, resulting in 17 independent frequency components (m=0,1,…,16), with the corresponding frequencies as follows:

[0064] ;

[0065] Right now , , ..., .

[0066] S2.2 Extract low-frequency features from the energy distribution of the full-band frequency components to obtain low-frequency feature sequences for each spatial partition.

[0067] The steps for extracting the low-frequency features include A1 to A4:

[0068] A1. Arrange the energy components below the cutoff frequency of dynamic occlusion change in the full-band frequency component energy distribution according to the frame order to obtain the time sequence of low-frequency component energy in each spatial partition, and store it as a low-frequency feature sequence in the occlusion evolution spectrum.

[0069] A cutoff frequency for dynamic changes in occlusion is set. This cutoff frequency is used to distinguish between the low-frequency components reflecting the slow spread of occlusion and the high-frequency components reflecting brief random disturbances in the occlusion occupancy time series. The value of the cutoff frequency for dynamic changes in occlusion is determined based on the physical characteristics of occlusion spread: occlusion typically takes several seconds to tens of seconds to spread from the edge of the field of view to the center, corresponding to a frequency much lower than the high-frequency disturbances introduced by randomly moving targets. Therefore, the cutoff frequency for dynamic changes in occlusion can be set to a value in the range of 1 to 2 Hz.

[0070] Let the set of frequency component indices corresponding to the low-frequency range be . , It is the dynamically changing cutoff frequency of the blockage, where m is an independent frequency component. It corresponds to the frequency.

[0071] For each spatial partition (i,j) at each short-time window frame position, sum the energy of the low-frequency components in the full-band frequency component energy distribution to obtain the low-frequency component energy of that frame: ; It is the energy distribution of frequency components across the entire frequency band.

[0072] Arranged by frame, the low-frequency component energy timing sequence of each spatial partition is obtained.

[0073] In this embodiment, the cutoff frequency is set to 1.5Hz, then the set of frequency component indices corresponding to the low-frequency range is as follows: All values ​​do not exceed the range after approximating the cutoff frequency for dynamic changes in occlusion, or m=1 according to the actual setting.

[0074] by For example, the low-frequency component energy of partition (2,3) at the first short time window is: .

[0075] A2. Normalize the energy distribution of the full-band frequency components within each short time window according to the total energy of the full-band to obtain the normalized energy distribution of the frequency components within each short time window.

[0076] For the full-band frequency component energy distribution of each spatial partition (i,j) within each short-time window kkk, normalize it according to the total full-band energy within that short-time window. This eliminates the incomparability of energy magnitudes caused by differences in the absolute value of occupancy between different short-time windows and different partitions, resulting in the normalized frequency component energy distribution:

[0077] ;

[0078] After normalization, It reflects the relative proportion of the energy of each frequency component in the total energy of the entire frequency band, and its value range is [0,1], and the sum of the normalized energy of each frequency component is 1.

[0079] Taking the first short-time window of partition (2,3) as an example, the total energy of the entire frequency band is:

[0080] ;

[0081] The normalized energy of the low-frequency components (m=1, 2) totals 38.6% + 28.2% = 66.8%, indicating that the energy of this partition is significantly concentrated in the low-frequency band within this time window, which is consistent with the low-frequency dynamic characteristics of occlusion spread.

[0082] A3. The centroid position of each frequency component in the low-frequency range of the normalized frequency component energy distribution is obtained by frequency weighting, thus obtaining the low-frequency centroid position sequence of each frame in each spatial partition.

[0083] For the normalized frequency component energy distribution of each spatial partition (i,j) at each short time window frame position, the normalized energy of each frequency component is weighted according to its corresponding frequency value in the low frequency range to obtain the centroid position (i.e., the frequency-weighted mean), thus obtaining the low-frequency centroid position:

[0084] ;

[0085] Low-frequency centroid position This reflects the location of the low-frequency energy centroid on the frequency axis within the current short time window: when the occlusion spread rate increases, the proportion of high-frequency components in the occlusion occupancy time sequence rises, and the low-frequency centroid shifts towards higher frequencies; conversely, when the occlusion tends to stabilize, energy concentrates at lower frequencies, and the centroid shifts towards lower frequencies. The low-frequency centroid position sequence is obtained by arranging the frames.

[0086] Using the first short time window of partition (2,3) and the low frequency range For example:

[0087] .

[0088] After all calculations are completed in the same way, the centroid position decreases with each frame, reflecting that low-frequency energy gradually concentrates to lower frequencies during the occlusion spread process of this partition, that is, the occlusion spread rhythm tends to be a slower and more stable dynamic mode.

[0089] A4. Differentiate the low-frequency centroid position sequence between adjacent frames to obtain the low-frequency centroid drift sequence of each spatial partition, and store the low-frequency centroid drift sequence together in the occlusion evolution spectrum.

[0090] For the low-frequency centroid position sequence of each spatial partition (i,j), in adjacent frames and By performing a first-order difference between them, we obtain the low-frequency centroid drift sequence:

[0091] ;

[0092] Where K is the total number of short time windows. The low-frequency centroid drift sequence reflects the direction and magnitude of the frequency axis displacement of the low-frequency energy centroid between adjacent frames: positive values ​​indicate that the centroid drifts towards higher frequencies (corresponding to increased occlusion dynamics), negative values ​​indicate that the centroid drifts towards lower frequencies (corresponding to decreased occlusion dynamics), and zero values ​​indicate that the low-frequency occlusion dynamics remain stable.

[0093] The low-frequency centroid drift sequence, as the second part of the low-frequency feature sequence, together with the low-frequency component energy time series obtained in step A1, constitutes the complete low-frequency feature sequence and is stored in the occlusion evolution spectrum. Compared with the centroid position sequence itself, the low-frequency centroid drift sequence is more sensitive to short-term changes in occlusion dynamics and can capture acceleration or deceleration inflection points during the occlusion propagation process.

[0094] S2.3 Construct an occlusion evolution spectrum from the structured records that use spatial partitions as row indexes and the low-frequency feature sequences as row content.

[0095] The low-frequency feature sequences of all spatial partitions are organized in a structured manner to construct an occlusion evolution spectrum. The occlusion evolution spectrum uses spatial partition (i,j) as the row index and the low-frequency feature sequence of each partition as the row content, forming the following structured record:

[0096] Each row corresponds to a spatial partition, containing two types of low-frequency feature sequences: the first is the low-frequency component energy time series obtained in step A1, which records the energy intensity of the low-frequency occlusion dynamics at each frame position of the partition; the second is the low-frequency centroid drift sequence obtained in step A4, which records the drift dynamics of the low-frequency energy centroid of the partition between adjacent frames. The two together constitute the low-frequency feature sequence content of the partition row.

[0097] S3. Perform cross-regional temporal correlation calculation on the low-frequency energy time series of each spatial partition in the occlusion evolution spectrum to obtain a cross-regional temporal correlation matrix group. Based on the temporal correlation magnitude and spatial adjacency relationship of each partition pair in the cross-regional temporal correlation matrix group, merge the spatial partitions to obtain the topology map of the occlusion spread area.

[0098] Specifically, the steps for performing the cross-partition time series correlation calculation include:

[0099] S3.1. For the low-frequency component energy time series of any two spatial partitions in the occlusion evolution spectrum, perform cross-partition time series correlation measurement within the allowable time lag range to obtain the time series correlation magnitude and partition pair time lag amount of each partition pair.

[0100] The steps for performing the cross-partition time-series correlation measurement include B1 to B6:

[0101] B1. For the low-frequency component energy time series of any two spatial partitions in the occlusion evolution spectrum, calculate the normalized cross-correlation function within the allowable time lag range to obtain the normalized cross-correlation function curve of each partition based on the low-frequency component energy time series, and extract the first correlation coefficient and the first time lag at the peak value.

[0102] For any two spatial partitions in the occlusion evolution spectrum and Take the energy timing of their respective low-frequency components. and Calculate the normalized cross-correlation function within the allowable time lag range.

[0103] Allowable time lag refers to the maximum frame offset allowed between two time series when calculating the cross-correlation function. This parameter is introduced to capture the temporal propagation relationship of occlusion spreading from one spatial partition to another: if there is a temporal propagation delay between the two partitions, the cross-correlation function will peak at the corresponding hysteresis, rather than reaching its maximum value at zero hysteresis. The maximum frame offset is determined based on the field of view spatial range and the typical rate of occlusion spread to ensure that the search is performed within the physically possible maximum propagation delay.

[0104] For a low-frequency component energy timing sequence with a timing length of K, in terms of hysteresis At this point, the normalized cross-correlation function is defined as:

[0105] ;

[0106] in, This represents the energy time series of the low-frequency components after mean removal. The normalized cross-correlation function has a range of [−1,+1], where a value of 1 indicates that the two sequences are perfectly positively correlated at that lag, a value of -1 indicates that they are perfectly negatively correlated, and a value of 0 indicates that they are not correlated.

[0107] right Through iterative calculations, the normalized cross-correlation function curve is obtained. The function value corresponding to the peak value is taken as the first correlation coefficient. The lag corresponding to the peak value is taken as the first time lag. :

[0108] ;

[0109] First time delay A positive sign indicates a partition. Low-frequency energy changes lead the partition in time. That is, the direction of the obstruction and spread is from point to Negative values ​​are the opposite.

[0110] By pairwise combining all spatial partitions in the occlusion evolution spectrum and repeating the above calculation, the first correlation coefficient of all partition pairs can be obtained. With the first time delay .

[0111] The field of view is divided into 16 spatial partitions, and the low-frequency component energy timing length K=11 (corresponding to 11 short time window frame positions). The allowable time lag range is defined. Each short time window step corresponds to 24 frames.

[0112] Taking partitions (2,3) and (2,4) as examples, the normalized cross-correlation function values ​​of the low-frequency component energy time series of the two are calculated at τ∈{−3,−2,−1,0,+1,+2,+3}, and the cross-correlation function curves are obtained.

[0113] Suppose that the peak value is obtained at τ=+1, then the first correlation coefficient First time delay This indicates that the low-frequency energy changes in partition (2,3) lead partition (2,4) by about one short time window step (8 frames) in time, meaning that the occlusion spreads from (2,3) to (2,4).

[0114] B2. For the low-frequency centroid drift sequences of any two spatial partitions in the occlusion evolution spectrum, calculate the normalized cross-correlation function within the allowable time lag range to obtain the normalized cross-correlation function curves of each partition based on the low-frequency centroid drift sequences, and extract the second correlation coefficient and the second time lag at the peak value.

[0115] For any two spatial partitions, calculate the normalized cross-correlation function curves of the low-frequency centroid drift sequence within the allowable time lag range. Extract the second correlation coefficient at the peak value. With the second time delay :

[0116] ;

[0117] Repeat the calculation for all partition pairs to obtain the second correlation coefficient and second time delay for all partition pairs.

[0118] B3. Sum the low-frequency component energy time series of each spatial partition on the time axis to obtain the total low-frequency component energy of each spatial partition. The proportion of the total low-frequency component energy in the total low-frequency component energy of all spatial partitions is used as the energy dominance.

[0119] Energy timing of low-frequency components for each spatial partition (i,j) The total low-frequency component energy of the partition is obtained by summing the positions of all frames on the time axis over the entire observation period.

[0120] The energy dominance of a region is determined by the proportion of the total low-frequency component energy W(i,j) of each region to the sum of the total low-frequency component energy of all spatial regions.

[0121] Energy dominance reflects the contribution of each spatial partition to the low-frequency occlusion dynamics during the entire field-of-view occlusion evolution process: partitions with higher energy dominance have stronger occlusion evolution signals, and their corresponding correlation coefficients and time delay estimates are more reliable. Therefore, they should be given higher weights in subsequent weighted fusion. The sum of the energy dominance of all spatial partitions is 1.

[0122] Suppose that the total low-frequency component energy of the 16 spatial partitions, after summing, is W(2,3) = 1.824 for partition (2,3), W(2,4) = 1.563 for partition (2,4), and the sum of the total energy of all 16 partitions is 12.480. Then:

[0123] The energy dominance α(2,3) = 1.824 / 12.480 = 0.146, α(2,4) = 1.563 / 12.480 = 0.125. The mean energy dominance of the two partitions is (0.146 + 0.125) / 2 = 0.136.

[0124] B4. The first correlation coefficient based on the energy time series of low-frequency components and the second correlation coefficient based on the centroid drift sequence are weighted and summed according to the mean energy dominance of the spatial partitions at both ends, respectively, to obtain the comprehensive time series correlation magnitude of each partition pair.

[0125] For any partition pair (p,q), its B1-derived first correlation coefficient The second correlation coefficient obtained with B2 These figures reflect the temporal correlation based on two types of low-frequency feature sequences. To integrate them into a unified comprehensive temporal correlation magnitude, the mean energy dominance of the corresponding partition on both spatial partitions is used. As the weights for each partition pair, the first and second correlation coefficients of all partition pairs are weighted and summed to obtain the overall time-series correlation magnitude. :

[0126] ;

[0127] The magnitude of the overall temporal correlation is simply the mean of the first and second correlation coefficients. The mean of energy dominance serves to weight the relative importance of the overall magnitudes between different partition pairs across the partition scale, ensuring that the correlations of partition pairs with high energy dominance have a greater influence on subsequent merge decisions. When there are many partition pairs and significant differences in energy dominance among partitions, this weighting mechanism can effectively suppress the interference of noise-dominated spurious correlations in low-energy partition pairs on the merge results.

[0128] Taking the partition pair (2,3) and (2,4) as an example, the first correlation coefficient is 0.91, the second correlation coefficient is 0.86, and the overall time series correlation magnitude is (0.91+0.86) / 2=0.885.

[0129] B5. Perform a sign consistency check on the first and second time delays:

[0130] If the two signs are the same, then the average energy dominance of the two spatial partitions at the corresponding partitions is weighted and summed to obtain the time delay of each partition.

[0131] The fact that both sequences have the same sign indicates that they agree on the direction of spread. Therefore, the two time delay quantities are directly weighted and summed using the mean of the energy dominance of the corresponding partition to both spatial partitions, yielding the partition-paired time delay quantity. :

[0132] ;

[0133] in and These are the mean energy dominance values ​​at both ends of the partition corresponding to the energy time series and centroid drift sequence, respectively. In this step, both are the same, both being [value missing]. Therefore, the above formula simplifies to the equal weighted average of the two time delays.

[0134] If the signs of the two are opposite, the sign of the time delay corresponding to the side with the higher mean of energy dominance of the two spatial partitions is used as the benchmark. The signs of the other side's time delay are then aligned, and then weighted and summed according to the mean of energy dominance to obtain the time delay of each partition pair.

[0135] The opposite signs of the two sequences indicate a contradiction in their judgments regarding the direction of spread. In this case, the sign of the time delay corresponding to the higher mean of the energy dominance of the two spatial partitions is used as the benchmark: Let... Then Based on the symbol, The symbols are aligned with the symbols. The same applies, and the time delay of the partition is obtained by weighting and summing the values ​​according to the mean of the energy dominance.

[0136] S3.2 Construct a time-series correlation magnitude matrix based on the time-series correlation magnitude of each partition pair, and construct a time-delay matrix based on the time-delay magnitude of each partition pair.

[0137] Construct a time series correlation magnitude matrix based on the comprehensive time series correlation magnitude of each partition pair. A time delay matrix T is constructed based on the time delay of each partition. Both matrices are P×P symmetric matrices (P is the total number of spatial partitions), and the matrix elements (p,q) correspond to the magnitude of the comprehensive temporal correlation and the time delay between partition p and partition q, respectively. The diagonal elements (the correlation between a partition and itself) are set to 1 (correlation magnitude matrix) and 0 (time delay matrix).

[0138] S3.3 Record the time series correlation magnitude matrix and the time delay matrix together as a cross-partition time series correlation matrix group.

[0139] It should also be noted that the steps for merging the spatial partitions include S3.4 to S3.8:

[0140] S3.4 Construct a spatial adjacency graph using spatial partitions as nodes and spatial adjacency relationships as hard connection constraints.

[0141] A spatial adjacency graph G is constructed using all spatial partitions as nodes and spatial adjacency relationships as hard connection constraints. Spatial adjacency relationship refers to the geometric relationship in the field of view where two spatial partitions are directly adjacent in spatial location and share a common boundary.

[0142] The four-neighbor connection rule is adopted: a connection edge is established between partition (i,j) and the partitions above, below, to the left and to the right, and no connection edge is established between non-adjacent partitions.

[0143] S3.5 For each adjacent partition pair in the spatial adjacency graph, extract the corresponding edge weights from the temporal correlation magnitude matrix of the cross-partition temporal correlation matrix group to obtain a spatial adjacency weighted graph.

[0144] For each connecting edge in the spatial adjacency graph G, i.e. each pair of adjacent partitions (p,q), the corresponding element is extracted from the temporal correlation magnitude matrix R of the cross-partition temporal correlation matrix group as the edge weight, thus obtaining the spatial adjacency weighted graph.

[0145] S3.6. Merge the connecting edges in the spatial adjacency weighted graph in descending order of edge weight, and merge the spatial partitions that are in the same connected domain after the gradual merging into a single occlusion spread region.

[0146] Spatial adjacency weighted graph All connecting edges in the array are sorted from highest to lowest weight, and processed sequentially using a step-by-step merging strategy: each time, the connecting edge with the highest weight is selected, and the connected components containing the nodes at both ends of that edge are merged into the same connected component; the above process is repeated until a preset stopping condition is met (e.g., the weight of the remaining unmerged edges is lower than a preset relevance threshold). (or the number of connected components reaches the preset minimum number of occlusion spread areas).

[0147] After merging stops, all spatial partitions within the same connected component constitute an occlusion spread region, with different connected components corresponding to different occlusion spread regions. A preset correlation threshold is used. The threshold setting is determined based on the typical synchronization level of field-of-view occlusion spread: if the threshold is too high, it is easy to split the partition of the same spread process into multiple regions; if the threshold is too low, it is easy to mistakenly merge independent occlusion regions.

[0148] Set a correlation threshold The 24 connecting edges in the spatial adjacency weighted graph are sorted from highest to lowest weight and then merged step by step. First, the partitions with the highest weight (2,3) and (2,4) (weight 0.885) are merged, then partitions (2,2) and (2,3) (weight 0.871) are merged, then partitions (2,4) and (3,4) (weight 0.823) are merged, and so on, until the remaining edge weights are all below 0.70.

[0149] Ultimately, we arrive at the assumption that there are three occlusion spread regions: occlusion spread region It contains five spatial partitions: (2,2), (2,3), (2,4), (3,3), and (3,4), which constitute the main shading spread area; the shading spread area It contains two spatial partitions, (1,3) and (1,4); shading and spread area. It includes the remaining 9 spatial partitions, corresponding to the background areas with weaker occlusion.

[0150] S3.7. For each shading spread area, the time delay is weighted according to the time-series correlation magnitude and the average is taken to obtain the representative time delay within each shading spread area; for the partitions that cross the boundaries of different shading spread areas, the time delay is weighted according to the time-series correlation magnitude and the average is taken to obtain the representative time delay between each shading spread area.

[0151] For each area of ​​obstruction and spread All internal partition pairs (p,q) and ), with the corresponding elements in the time delay matrix T The time delay is represented by the corresponding element in the time-series correlation magnitude matrix R. Using the weights as values, a weighted average is performed to obtain the representative time delay within the shading spread area. :

[0152] ;

[0153] The representative time delay within the area reflects the average propagation delay of the shading and spread between different zones within the spread area, and can be used as an internal consistency indicator of the shading and spread rhythm in the area.

[0154] For spreading across different shielding areas and All partition pairs (p,q) at the boundary (And p and q have a connecting edge in the spatial adjacency graph). Using the corresponding elements in the time delay matrix T as the time delay and the corresponding elements in the temporal correlation magnitude matrix R as the weights, a weighted average is performed to obtain the occlusion spread area. and The interval between them represents the time delay. :

[0155] ;

[0156] The sign of the interval representing the time delay reflects the direction of occlusion propagation between two occlusion spread regions: a positive value indicates... Leading in time sequence That is, the occlusion is caused by Towards Directional spread; negative values ​​indicate the opposite.

[0157] To shield the spread area (including partitions (2,2), (2,3), (2,4), (3,3), (3,4)) and (Taking partitions (1,3) and (1,4) as an example, spanning...) and The adjacent partition pairs of the boundary include (2,3) and (1,3) (edge ​​weight R=0.741, time delay T=+2) and (2,4) and (1,4) (edge ​​weight R=0.712, time delay T=+2), then:

[0158] ;

[0159] The interval represents a time delay of +2, indicating that the occlusion is caused by... Towards The propagation direction is spread, and the propagation delay is approximately two short time window steps (16 frames).

[0160] S3.8. Based on the time delay represented by the interval, perform the first encoding on the propagation time sequence relationship between all occlusion spread areas to obtain the topology map of the occlusion spread area.

[0161] The steps for performing the first encoding include C1 to C4:

[0162] C1. Based on the positive and negative directions of the time delay represented by the interval, assign the propagation direction attribute to the connecting edges between each occlusion propagation region to obtain a directed topology graph.

[0163] The time delay is represented by the intervals between each pair of the entire shading spread area. Based on their positive or negative sign, the connecting edges between each occlusion propagation area are assigned a propagation direction attribute: if Then the direction of the connecting edge is from point to For each pair of occlusion-spreading regions, only one directed edge is retained. A directed topology graph is constructed from all occlusion-spreading regions and their directed connecting edges. .

[0164] With 3 shielding spread areas For example, , ,Right now Leading), set The directed topological graph is then: ,at the same time This forms a directed acyclic graph structure, reflecting the occlusion from... Starting from there, it spread to... And then spread further to The propagation chain.

[0165] C2. For the longest directed path in the directed topology graph that starts from each shading spread area and reaches along the propagation direction, the sum of the total number of spatial partitions covered by all shading spread areas on the path is taken as the path spatial span.

[0166] For each shading spread region in the directed topology graph The longest directed path from the starting point along the direction of propagation (i.e., from the point in the graph) The directed path with the most reachable nodes is used as the path's spatial span, which is the sum of the total number of spatial partitions covered by all shading and sprawl areas along that path. The path spatial span comprehensively reflects the occlusion from... The number of zones indicates the maximum spatial range that can be spread from the starting point. The more zones there are, the larger the spatial span, reflecting the more important the source position of the occlusion spread area in the entire field of view spread process.

[0167] C3. The ratio of the path spatial span to the cumulative value of the interval representative time delay of each pair of adjacent shading spread areas on the longest directed path is used as the spread rate of each shading spread area.

[0168] Based on path space span The ratio of the cumulative interval representative time delay of each adjacent shading spread region pair on the longest directed path to the spread rate of each shading spread region is used as the spread rate of each shading spread region. :

[0169] ;

[0170] in, Indicates The set of all adjacent occlusion propagation pairs on the longest directed path starting from point A. This represents the absolute value of the time delay in the corresponding interval.

[0171] by Starting from point A, the longest directed path is The cumulative time delay value of adjacent pairs of regions is If the path space span is 16, then:

[0172] ;

[0173] by Starting from point A, the longest directed path is Given a cumulative time delay of 2.0 and a path space span of 7, then:

[0174] ;

[0175] No subsequent path .

[0176] C4. Normalize the spread rate by the maximum value of the spread rate of all shading spread areas to obtain the normalized spread rate, and record the normalized spread rate in each corresponding shading spread area node in the topology graph of the shading spread area.

[0177] The spread rate of the entire shielded spread area The maximum spread rate among all the blocked spread areas. Normalization is performed to obtain the normalized spread rate of each shading spread region. :

[0178] ;

[0179] The normalized spread rate ranges from (0,1). The closer the value is to 1, the faster the occlusion spreads in the field of view, and the higher the dynamic risk level should be assigned in failure assessment. The normalized spread rate is recorded in the corresponding occlusion spread area nodes in the occlusion spread area topology map to complete the first encoding, and finally the occlusion spread area topology map carrying the propagation direction attribute and the normalized spread rate node markers is obtained.

[0180] Specifically, the spread rates of all three shielding spread zones are as follows: The maximum value is 3.5 after normalization:

[0181] ;

[0182] The above-mentioned normalized spread rates are recorded in the topology map of the shading spread region. , , The node attributes are as follows. At this point, the topology of the occlusion spread region is completed. Its nodes are 3 occlusion spread regions. The node attributes include the set of spatial partitions covered by each region and the normalized spread rate. The directed edges carry the interval representative time delay and propagation direction attributes, which fully describe the spatial topology and dynamic propagation law of occlusion spreading outward from the origin region within the field of view.

[0183] S4. Normalize the spatial coverage of the occlusion spread area in the topology map and the low-frequency energy ratio of the corresponding spatial partition in the occlusion evolution spectrum. Based on the combined magnitude of the normalized spatial coverage and low-frequency energy ratio, determine the field of view failure status and output the field of view failure determination result.

[0184] Specifically, the steps for determining the field-of-view failure state include S4.1 to S4.7:

[0185] S4.1 Summing the low-frequency component energy time series of each spatial partition in the occlusion evolution spectrum on the time axis to obtain the total low-frequency component energy of each spatial partition.

[0186] For the low-frequency component energy time series of each spatial partition (i,j) in the occlusion evolution spectrum, the summation is performed on the time axis over all short time window frames to obtain the total accumulated low-frequency component energy of that partition during the entire observation period: .

[0187] Total energy of low-frequency components This reflects the cumulative intensity of low-frequency dynamic components being obscured by each spatial region throughout the entire observation period. A larger value indicates a deeper and longer-lasting impact of occlusion spread on that region during the observation period; conversely, a smaller value indicates a lower value. A smaller value indicates that the shading fluctuations are mainly high-frequency random disturbances, with low-frequency spread components being insignificant.

[0188] The field of view is divided into 16 spatial partitions of 4×4, with a low-frequency component energy timing length K=11. Typical calculation results for the total low-frequency component energy in each spatial partition are as follows:

[0189] The low-frequency component energy time series of partition (2,3) is {0.147,0.163,0.201,0.258,0.312,0.298,0.271,0.243,0.218,0.196,0.181}, and the summation result is 2.488.

[0190] The low-frequency component energy time series of partition (2,4) is {0.112,0.138,0.179,0.231,0.287,0.271,0.248,0.219,0.197,0.176,0.162}, and the summation result is 2.220.

[0191] The low-frequency component energy time series of partition (1,1) is of a relatively low magnitude, and the summation result is 0.341, which reflects that the partition blocks the low-frequency dynamic components weakly.

[0192] S4.2. Normalize the total energy of the low-frequency components according to the total energy of the partitioned occlusion fluctuation sequence on the time axis to obtain the low-frequency energy ratio of each spatial partition.

[0193] The total low-frequency component energy of each spatial partition is the energy accumulated in the frequency band below the low-frequency cutoff frequency, and its absolute value is affected by the overall energy level of the occupancy wave sequence of that partition. To eliminate the incomparability of magnitudes caused by the difference in the absolute value of occupancy between different partitions, the total low-frequency component energy needs to be normalized according to the total energy of the occupancy wave sequence of that partition on the time axis.

[0194] The total energy of the occlusion wave sequence in partition (i,j) is defined as:

[0195] ;

[0196] in, This is the partitioned occlusion wave sequence obtained in step S1. The low-frequency energy proportion of each spatial partition. Defined as: .

[0197] The proportion of low-frequency energy reflects the relative weight of low-frequency dynamic components in the full-band wave energy of the occlusion wave sequence. The higher the value, the more the shading fluctuations in that region manifest as slowly spreading low-frequency dynamics, rather than high-frequency components of brief random disturbances. The lower the value, the more likely the shading fluctuations in that region are dominated by random high-frequency disturbances, and the lower the correlation with shading propagation failure. Using the proportion of low-frequency energy rather than the absolute amount of low-frequency energy as the subsequent judgment criterion can effectively avoid misjudging regions with a high absolute value of shading occupancy but dominated by random disturbances.

[0198] Taking partition (2,3) as an example, its occlusion fluctuation sequence The total energy at T=100 frames is obtained by summing the squares of each frame. Then the proportion of low-frequency energy is:

[0199] ;

[0200] Taking partition (1,1) as an example, the total energy of its occlusion wave sequence is 0.682, and the total energy of the low-frequency components is 0.341. .

[0201] The proportion of low-frequency energy in partition (2,3) is much higher than that in partition (1,1), indicating that the low-frequency propagation component in the shading fluctuation is more significant in the former and is more likely to be in the core area of ​​the shading propagation path.

[0202] S4.3. Normalize the low-frequency energy ratio according to the maximum value of the low-frequency energy ratio of all spatial partitions to obtain the normalized low-frequency energy of each spatial partition.

[0203] The proportion of low-frequency energy in all spatial zones is determined by the maximum proportion of low-frequency energy in all spatial zones. Normalization is performed to obtain the normalized low-frequency energy of each spatial partition. :

[0204] .

[0205] The normalized value of low-frequency energy ranges from (0,1], and the spatial partition with the maximum value of 1 is the partition with the most prominent low-frequency occlusion dynamics among all partitions. By normalizing the low-frequency energy proportion to the maximum value, the low-frequency dynamic intensity of each partition is mapped to a unified dimension, eliminating the difference in the absolute magnitude of the low-frequency energy proportion between different video scenes.

[0206] S4.4. The low-frequency energy normalization of all spatial partitions in each shading spread area in the topology map of the shading spread area is aggregated by mean according to the number of spatial partitions to obtain the representative low-frequency energy of each shading spread area.

[0207] For each shading-spread region in the topology map, the normalized low-frequency energy of all spatial partitions it contains is taken as the arithmetic mean by the number of spatial partitions to obtain the representative low-frequency energy of that shading-spread region. :

[0208] ;

[0209] in, This represents the number of spatial zones contained within the shading spread area. The low-frequency energy representative value reflects the average level of low-frequency shading dynamic intensity of each spatial zone within the shading spread area, serving as a representative indicator of the overall shading spread intensity of the area. The purpose of using mean aggregation instead of summation is to eliminate the influence of the size of the shading spread area on the energy representative value, ensuring comparability of low-frequency energy representative values ​​between shading spread areas of different areas, and preventing the artificial inflation of the energy representative value of large-area spread areas due to the inclusion of more low-intensity zones.

[0210] S4.5. Normalize the spatial coverage range of each occlusion spread area in the topology map of the occlusion spread area according to the total number of field of view partitions to obtain the spatial coverage normalization amount.

[0211] For each occlusion spread region in the topology map, its spatial coverage is represented by the number of spatial partitions it contains. This coverage is then normalized to the total number of field-of-view partitions P (P=16 in this example) to obtain the normalized spatial coverage of each occlusion spread region:

[0212] ;

[0213] Spatial Coverage Normalization It reflects the spatial proportion of the occupancy area in the entire field of view, and its value ranges from (0,1). The larger the value, the wider the field of view covered by the occlusion spread area, and the greater its impact on the effective observation capability of the field of view. When the spatial coverage normalization of a certain occlusion spread area is close to 1, it indicates that the occlusion has spread to most of the field of view, and the field of view tends to be completely ineffective.

[0214] S4.6. The average spatial center coordinates of all spatial partitions within each occlusion spread area are used as the regional center coordinates of the occlusion spread area. The first distance between the regional center coordinates and the field of view center coordinates is calculated. The first distance of each occlusion spread area is normalized by the maximum value of the first distance among all occlusion spread areas. The complement of the normalized first distance is used as the positional importance of each occlusion spread area.

[0215] The central region of the field of view is generally of higher importance for capturing monitored targets. The closer the obstruction spread area is to the center of the field of view, the greater its impact on the effective observation capability of the field of view, and it should be given higher weight in failure judgment. Therefore, the positional importance of each obstruction spread area is defined by its distance relationship from the center of the field of view.

[0216] Specifically, the average spatial center coordinates of all spatial partitions within the occlusion spread area are used as the regional center coordinates of that area. The spatial center coordinates of each spatial partition (i,j) are defined as the geometric center position of that partition in the image coordinate system. Then the area of ​​spread will be blocked. Regional center coordinates for:

[0217] ;

[0218] Field center coordinates Defined as the geometric center position of the entire field of view. Calculate the Euclidean distance between the coordinates of the region center and the coordinates of the field of view center, and use this as the first distance:

[0219] ;

[0220] The maximum value of the first distance in the entire occlusion spread area. Normalize the first distance of each shading spread area to obtain the normalized first distance. Then, its complement value is used as the positional importance of each occlusion spread area:

[0221] ;

[0222] Location importance The value range is [0,1]. The closer the center of the region is to the center of the field of view, the better. The smaller, The larger the value, the higher the positional importance. The positional importance of the occlusion spread area farthest from the center of the field of view is 0, and the positional importance of the occlusion spread area closest to the center of the field of view is 1 (the maximum value of 1 is obtained only when the center of the area coincides with the center of the field of view).

[0223] S4.7. The normalized spatial coverage, the low-frequency energy representative quantity, and the normalized spread rate are weighted and accumulated according to the location importance to obtain the comprehensive magnitude of field failure, and the field failure judgment result is output based on the comprehensive magnitude of field failure.

[0224] For each area of ​​shading and spread, the spatial coverage normalization is calculated based on its location importance. Low-frequency energy representation With normalized spread rate The three indicators are weighted and accumulated to obtain the comprehensive magnitude of the field of view failure: .

[0225] Overall magnitude of field of view failure The three dimensions of spatial coverage of the obstruction, low-frequency energy intensity, and spread speed are uniformly incorporated into the discrimination criteria, and the contribution of each obstruction spread area is differentiated and weighted by position importance: the obstruction spread area with high position importance (i.e., close to the center of the field of view) contributes more to the overall magnitude, reflecting the physical law that the obstruction in the center of the field of view poses a higher threat to the failure of the field of view.

[0226] Based on the comparison between the overall magnitude of field-of-view failure and the preset discrimination threshold, the field-of-view failure discrimination result is output. The discrimination threshold can be set according to the failure tolerance of the actual application scenario, for example, setting a three-level threshold: when When the field of view is determined to be normal, the output field of view is normal; when When the field of view is partially damaged, a field of view alarm is output, indicating that the field of view occlusion is spreading but has not yet completely failed; when When the field of view is determined to be completely failed, a field of view failure status is output, a failure alarm is triggered, and an emergency response procedure is initiated. and All of these are set threshold values.

[0227] Example 3 is an embodiment of the present invention, which provides a field-of-view failure discrimination system based on occlusion evolution spectrum, including:

[0228] The fluctuation sequence extraction unit is used to remove the DC component of the occlusion occupancy of each spatial partition within the field of view based on the frame mean, so as to obtain the partition occupancy fluctuation sequence.

[0229] The occlusion evolution spectrum construction unit is used to perform time-frequency joint decomposition on the partitioned occlusion fluctuation sequence to obtain the time spectrum, extract the distribution of low-frequency component energy of each spatial partition with the frame from the time spectrum, and construct the occlusion evolution spectrum.

[0230] The shading spread region topology map generation unit is used to perform cross-partition temporal correlation calculation on the low-frequency energy time series of each spatial partition in the shading evolution spectrum to obtain a cross-partition temporal correlation matrix group. Based on the temporal correlation magnitude and spatial adjacency relationship of each partition pair in the cross-partition temporal correlation matrix group, the spatial partitions are merged to obtain the shading spread region topology map.

[0231] The field of view failure discrimination unit is used to normalize the spatial coverage of the occlusion spread area in the topology map of the occlusion spread area and the low-frequency energy ratio of the corresponding spatial partition in the occlusion evolution spectrum, respectively, and to judge the field of view failure state based on the comprehensive magnitude of the normalized spatial coverage area and the low-frequency energy ratio, and output the field of view failure discrimination result.

[0232] This embodiment also provides a computer device applicable to a field-of-view failure discrimination method based on occlusion evolution spectrum, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the field-of-view failure discrimination method based on occlusion evolution spectrum proposed in the above embodiment.

[0233] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a field failure discrimination method based on occlusion evolution spectrum as proposed in the above embodiments.

[0234] The storage medium proposed in this embodiment and the implementation of a field failure discrimination method based on occlusion evolution spectrum proposed in the above embodiment belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0235] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory, random access memory, flash memory, hard disk, or optical disk, and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0236] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for determining field-of-view failure based on occlusion evolution spectrum, characterized in that, Includes the following steps: The occlusion occupancy of each spatial partition within the field of view is removed by removing the DC component based on the frame mean, resulting in a partition occlusion fluctuation sequence. The time-frequency joint decomposition of the partitioned occlusion fluctuation sequence is used to obtain the time spectrum. The distribution of low-frequency component energy of each spatial partition with frame is extracted from the time spectrum to construct the occlusion evolution spectrum. Cross-regional temporal correlation calculation is performed on the low-frequency energy time series of each spatial region in the occlusion evolution spectrum to obtain a cross-regional temporal correlation matrix group. Based on the temporal correlation magnitude and spatial adjacency relationship of each pair of regions in the cross-regional temporal correlation matrix group, the spatial regions are merged to obtain the topology map of the occlusion spread region. The spatial coverage area of ​​the occlusion spread area in the topology map and the low-frequency energy ratio of the corresponding spatial partition in the occlusion evolution spectrum are normalized respectively. The field of view failure status is judged based on the comprehensive magnitude of the normalized spatial coverage area and the low-frequency energy ratio, and the field of view failure judgment result is output.

2. The field-of-view failure determination method based on occlusion evolution spectrum as described in claim 1, characterized in that, The steps for performing the time-frequency joint decomposition include: The partition occupancy fluctuation sequence is segmented according to a preset frame length. A short time window function is applied to the occupancy occupancy time sequence in each segment. Frequency decomposition is performed on the occupancy occupancy time sequence in each short time window to obtain the full-band frequency component energy distribution of each spatial partition at each frame position. Low-frequency features are extracted from the energy distribution of the frequency components across the entire frequency band to obtain low-frequency feature sequences for each spatial partition. The structured records, with spatial partitions as row indexes and the low-frequency feature sequences as row content, are constructed into an occlusion evolution spectrum.

3. The field-of-view failure determination method based on occlusion evolution spectrum as described in claim 2, characterized in that, The steps for extracting the low-frequency features include: The energy components below the cutoff frequency of dynamic occlusion change in the full-band frequency component energy distribution are arranged by frame to obtain the time sequence of low-frequency component energy in each spatial partition, which is then stored as a low-frequency feature sequence in the occlusion evolution spectrum. The energy distribution of the full-band frequency components within each short time window is normalized according to the total energy of the full-band to obtain the normalized energy distribution of the frequency components within each short time window. The centroid positions of each frequency component in the low-frequency range of the normalized frequency component energy distribution are obtained by frequency weighting, thus obtaining the low-frequency centroid position sequence of each frame in each spatial partition. The low-frequency centroid position sequence is differentially divided between adjacent frames to obtain the low-frequency centroid drift sequence of each spatial partition, and the low-frequency centroid drift sequence is stored together in the occlusion evolution spectrum.

4. The field-of-view failure determination method based on occlusion evolution spectrum as described in claim 3, characterized in that, The steps for performing the cross-partition time series correlation calculation include: For the low-frequency component energy time series of any two spatial partitions in the occlusion evolution spectrum, cross-partition time series correlation is measured within the allowable time lag range to obtain the magnitude of time series correlation and time lag of each partition pair. Construct a time-series correlation magnitude matrix based on the time-series correlation magnitude of each partition pair, and construct a time-delay matrix based on the time-delay magnitude of each partition pair; The time-series correlation magnitude matrix and the time-delay matrix are recorded together as a cross-partition time-series correlation matrix group.

5. The field-of-view failure determination method based on occlusion evolution spectrum as described in claim 4, characterized in that, The steps for performing the cross-partition time-series correlation measurement include: For the low-frequency component energy time series of any two spatial partitions in the occlusion evolution spectrum, the normalized cross-correlation function is calculated within the allowable time lag range to obtain the normalized cross-correlation function curve of each partition based on the low-frequency component energy time series, and the first correlation coefficient and the first time lag at the peak are extracted from it. For the low-frequency centroid drift sequences of any two spatial partitions in the occlusion evolution spectrum, the normalized cross-correlation function is calculated within the allowable time lag range to obtain the normalized cross-correlation function curves of each partition based on the low-frequency centroid drift sequences, and the second correlation coefficient and the second time lag at the peak are extracted from them. The energy time series of low-frequency components in each spatial partition are summed on the time axis to obtain the total energy of low-frequency components in each spatial partition. The proportion of the total energy of low-frequency components in the total energy of low-frequency components in all spatial partitions is used as the energy dominance. The first correlation coefficient based on the energy time series of low-frequency components and the second correlation coefficient based on the centroid drift sequence are weighted and summed according to the mean energy dominance of the two spatial partitions at the corresponding partitions to obtain the comprehensive time series correlation magnitude of each partition pair. Perform sign consistency checks on the first and second time delay quantities: If the two signs are the same, then the mean values ​​of energy dominance of the two spatial partitions at the corresponding partitions are weighted and summed to obtain the time delay of each partition. If the signs of the two are opposite, the sign of the time delay corresponding to the side with the higher mean of energy dominance of the two spatial partitions is used as the benchmark. The signs of the other side's time delay are then aligned, and then weighted and summed according to the mean of energy dominance to obtain the time delay of each partition pair.

6. The field-of-view failure determination method based on occlusion evolution spectrum as described in claim 5, characterized in that, The steps for merging the spatial partitions include: A spatial adjacency graph is constructed using spatial partitions as nodes and spatial adjacency relationships as hard connection constraints. For each adjacent partition pair in the spatial adjacency graph, the corresponding edge weights are extracted from the temporal correlation magnitude matrix of the cross-partition temporal correlation matrix group to obtain a spatial adjacency weighted graph. The connecting edges in the spatial adjacency weighted graph are gradually merged according to the edge weight level from high to low, and the spatial partitions that are in the same connected domain after gradual merging are merged into a single occlusion spread area. The time delay values ​​within each shading spread area are weighted according to the magnitude of temporal correlation and averaged to obtain the representative time delay value within each shading spread area; the time delay values ​​across the boundaries of different shading spread areas are weighted according to the magnitude of temporal correlation and averaged to obtain the representative time delay value between each shading spread area. Based on the time delay represented by the interval, the propagation time sequence relationship between all occlusion spread areas is first encoded to obtain the topology map of the occlusion spread area.

7. The field-of-view failure determination method based on occlusion evolution spectrum as described in claim 6, characterized in that, The steps for performing the first encoding include: Based on the positive and negative directions of the time delay represented by the interval, the connecting edges between each occlusion spread region are assigned propagation direction attributes to obtain a directed topology graph. For the longest directed path in the directed topology graph that starts from each shading spread area and reaches along the propagation direction, the sum of the total number of spatial partitions covered by all shading spread areas on the path is taken as the path spatial span. The spread rate of each shading spread area is defined as the ratio of the path spatial span to the cumulative value of the interval representative time delay of each pair of adjacent shading spread areas on the longest directed path. The spread rate is normalized to the maximum value of the spread rates of all shading spread areas to obtain the normalized spread rate, and the normalized spread rate is recorded in each corresponding shading spread area node in the topology graph of the shading spread area.

8. The field-of-view failure determination method based on occlusion evolution spectrum as described in claim 7, characterized in that, The steps for determining the field of view failure state include: The low-frequency component energy time series of each spatial partition in the occlusion evolution spectrum are summed on the time axis to obtain the total low-frequency component energy of each spatial partition; The total energy of the low-frequency components is normalized according to the total energy of the partitioned occlusion fluctuation sequence on the time axis to obtain the low-frequency energy ratio of each spatial partition. The low-frequency energy ratio is normalized to the maximum value of the low-frequency energy ratio of all spatial partitions to obtain the normalized low-frequency energy of each spatial partition. The low-frequency energy normalization of all spatial partitions within each shading spread area in the topology map of the shading spread area is aggregated by mean according to the number of spatial partitions to obtain the representative low-frequency energy of each shading spread area. The spatial coverage range of each occlusion spread area in the topology map of the occlusion spread area is normalized according to the total number of field of view partitions to obtain the spatial coverage normalization amount. The average spatial center coordinates of all spatial partitions within each occlusion spread area are used as the regional center coordinates of that occlusion spread area. The first distance between the regional center coordinates and the field of view center coordinates is calculated. The first distance of each occlusion spread area is normalized by the maximum value of the first distance among all occlusion spread areas. The complement of the normalized first distance is used as the positional importance of each occlusion spread area. The normalized spatial coverage, the representative low-frequency energy, and the normalized spread rate are weighted and accumulated according to the location importance to obtain the comprehensive magnitude of field failure. The field failure judgment result is output based on the comprehensive magnitude of field failure.

9. A field-of-view failure discrimination system based on occlusion evolution spectrum, employing the field-of-view failure discrimination method based on occlusion evolution spectrum as described in any one of claims 1 to 8, characterized in that, include: The fluctuation sequence extraction unit is used to remove the DC component of the occlusion occupancy of each spatial partition within the field of view based on the frame mean, so as to obtain the partition occupancy fluctuation sequence. The occlusion evolution spectrum construction unit is used to perform time-frequency joint decomposition on the partitioned occlusion fluctuation sequence to obtain the time spectrum, extract the distribution of low-frequency component energy of each spatial partition with the frame from the time spectrum, and construct the occlusion evolution spectrum. The shading spread region topology map generation unit is used to perform cross-partition temporal correlation calculation on the low-frequency energy time series of each spatial partition in the shading evolution spectrum to obtain a cross-partition temporal correlation matrix group. Based on the temporal correlation magnitude and spatial adjacency relationship of each partition pair in the cross-partition temporal correlation matrix group, the spatial partitions are merged to obtain the shading spread region topology map. The field of view failure discrimination unit is used to normalize the spatial coverage of the occlusion spread area in the topology map of the occlusion spread area and the low-frequency energy ratio of the corresponding spatial partition in the occlusion evolution spectrum, respectively, and to judge the field of view failure state based on the comprehensive magnitude of the normalized spatial coverage area and the low-frequency energy ratio, and output the field of view failure discrimination result.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the field failure discrimination method based on the occlusion evolution spectrum according to any one of claims 1 to 8; It also includes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the field failure discrimination method based on the occlusion evolution spectrum according to any one of claims 1 to 8.

Images