Method and device for identifying logistics irregular behavior based on monitoring video

Abstract

The present application relates to the technical field of image data processing, in particular to a logistics irregular behavior identification method and device based on monitoring video, comprising: acquiring images in a logistics transportation process and extracting a plurality of shadow boundaries, using the midpoint of the shadow boundary as a seed point for region growing to generate an initial shadow; if the area ratio of the initial shadow to the shadow boundary is less than a set threshold, the initial shadow is removed to obtain a second shadow, and a behavior analysis algorithm is used to identify the behavior of the second shadow; the shadow boundary extraction specifically comprises: corroding the connected domain in the image foreground to obtain a first target, and obtaining a second target through subtraction operation, taking the geometric center of the second target as the center of a unit circle to obtain the points on the unit circle to form a feature vector; the feature vector is mapped to a three-dimensional space for clustering, and the mean value of the target value in the clustering cluster is calculated to obtain the shadow boundary. The present application solves the problem of low accuracy of logistics monitoring.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology. More specifically, this invention relates to a method and apparatus for identifying logistics violations based on surveillance video. Background Technology

[0002] With the rapid development of e-commerce, intelligent manufacturing, and modern warehousing and logistics, the scale of logistics operations is constantly expanding. Sorting, handling, loading / unloading, and stacking operations heavily rely on the collaboration of manual labor and equipment. In logistics parks, warehouses, and transfer centers, to ensure operational efficiency and safety, numerous surveillance cameras are typically deployed in the work area to monitor personnel actions, equipment operating status, and work processes. Existing logistics management systems mostly rely on manual inspections or post-event retrieval of surveillance video to verify and hold accountable those responsible for violations, such as improper throwing of goods, excessive stacking, deviation from designated routes, improper forklift operation, and failure to wear safety protective equipment. However, with the continuous increase in the complexity of logistics operations and the scale of video data, traditional regulatory methods relying on manual labor or simple rules are no longer sufficient to meet actual needs.

[0003] In existing technologies, one type of solution mainly relies on manual monitoring of surveillance videos in real time or random checks afterward. This approach not only requires a significant investment of manpower but is also limited by subjective experience and attention, making it prone to missed detections and misjudgments, and failing to promptly detect and stop logistics violations, exhibiting a significant lag. Another type of solution attempts to introduce automated recognition methods based on image processing or simple object detection algorithms, such as judging the presence of abnormal behavior by detecting the appearance features of personnel or goods.

[0004] However, such methods typically only identify a single type of violation, relying on pre-set fixed rules or thresholds, and are poorly adaptable to complex and diverse logistics violations. Furthermore, logistics operation scenarios often feature significant variations in lighting conditions, high personnel density, severe occlusion, and diverse operational processes. Existing technologies have limited capabilities in handling multi-target interactions, continuous action recognition, and behavioral temporal relationship modeling, easily misidentifying normal operational behaviors as violations, or failing to accurately distinguish the essential differences between similar actions. For example, existing technologies struggle to reliably differentiate between normal rapid handling and illegal throwing of goods, or between compliant temporary stacking and illegal excessive stacking, resulting in low accuracy in logistics monitoring. Summary of the Invention

[0005] To address the problem of low accuracy in logistics monitoring mentioned in the background art, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for identifying logistics violations based on surveillance video, comprising: acquiring images during logistics transportation and extracting multiple shadow boundaries from the images; using the midpoint of each shadow boundary as a seed point for region growing to obtain initial shadows; if the area ratio of the initial shadow to the corresponding shadow boundary is less than a set threshold, removing the initial shadow from the shadow boundary to obtain a second shadow, and using a behavior analysis algorithm to perform behavior recognition on the second shadow; the shadow boundary extraction process is as follows: performing erosion operations on each connected component in the foreground of the image to obtain a first target, and performing erosion operations on each connected component and the first target... The second target is obtained by performing a subtraction operation; points in different directions on the unit circle are obtained with the geometric center of the second target as the center, and the horizontal and vertical coordinates of the points and the target value are used to form a feature vector. The target value is inversely correlated with color uniformity and positively correlated with shadow probability. The color uniformity represents the stability of the direction, and the shadow probability represents the abnormality of the direction. The feature vectors of each point are mapped to points in three-dimensional space and clustered. The mean of the target value in each cluster is calculated, and the point corresponding to the feature vector in the cluster with the maximum mean is mapped to the boundary of the first target. The boundary with the longest continuity is obtained as the shadow boundary.

[0007] The aforementioned technical solution accurately distinguishes between initial and second shadows by judging region growth and shadow area ratio, eliminating areas that do not meet the conditions, avoiding false detections, and further identifying second shadows to improve the monitoring accuracy of dynamic objects. This solution optimizes the effects of shadow detection and dynamic monitoring, improves the accuracy and adaptability of intelligent monitoring systems, and is particularly effective in reducing false alarms and missed alarms in complex logistics and transportation environments, thereby enhancing monitoring accuracy.

[0008] Furthermore, the probability of the shadow is: In the formula, For the first The possible shadows in a given location, This represents the distance from the coordinates corresponding to the first color in the RGB color space to the point (1,1,1). This represents the distance from the coordinates corresponding to the first color in the RGB color space to the point (0,0,0). It is an exponential function with the natural constant e as its base. The maximum value among the RGB three channels of the first color. The first color is the minimum value among the three RGB channels of the first color; the first color is the RGB value that appears most frequently, and the RGB value is the average value of the three RGB channels of each intersection point. The intersection point is the point where a radial line drawn from the geometric center of the second target to a set direction intersects with the second target.

[0009] The aforementioned technical solution effectively improves the accuracy and reliability of shadow boundary extraction by precisely calculating shadow probability. By calculating radial lines starting from the geometric center of the target area, the boundary color at a given location can be accurately identified and constructed, thereby improving the sensitivity to color consistency and changes during shadow recognition. This method, by calculating shadow probability, helps distinguish different shadow types, especially in complex environments, allowing for more accurate judgment of shadow authenticity and type, avoiding the influence of interfering factors, and improving the accuracy of shadow detection and the ability to recognize dynamic objects in the monitoring system.

[0010] Furthermore, the first color is: a radial line is drawn with the geometric center of the second target as the center and directed to a set direction. The intersection point of the radial line and the second target is the target pixel point. The boundary color corresponding to the set direction is constructed based on the mean of the RGB three channels of the target pixel point. The points corresponding to the boundary colors of each set direction in the RGB color space are clustered to obtain multiple clusters. The boundary color corresponding to the largest cluster is taken as the first color of the set direction.

[0011] Furthermore, color uniformity is as follows: , For the first Color consistency for each set direction The number of boundary colors in the largest cluster. The number of connected components in the foreground of the image. The standard deviation of the distance from the point corresponding to each boundary color in the RGB color space to the center of the largest cluster is given. The standard deviation of the distance from the point corresponding to the boundary color of the largest cluster in the RGB color space to its cluster center. The th in the largest cluster The Euclidean distance from the point corresponding to the boundary color in the RGB color space to the center of its cluster.

[0012] The aforementioned technical solution effectively improves the accuracy and stability of shadow boundaries by calculating color uniformity. By analyzing the clustering and distribution characteristics of boundary colors, the consistency and dispersion of colors can be identified, thereby optimizing the localization and classification of shadow areas. This method comprehensively considers the color features in the largest cluster and the distance differences with other areas, which helps to accurately distinguish shadows in different areas and enhances adaptability to changes in illumination and color deviations. Overall, color uniformity calculation improves the accuracy of shadow detection and recognition, reduces misjudgments caused by environmental changes or noise interference, and thus improves the performance of the monitoring system in complex scenes.

[0013] Furthermore, the target value is: , For the first A target value with a set orientation. For the first The possible shadows in a given location, For the first The color uniformity of each set direction.

[0014] The aforementioned technical solution further improves the accuracy and stability of shadow detection by combining target values ​​of color uniformity and shadow probability. By integrating color uniformity and shadow probability into a comprehensive target value, a more refined judgment of the authenticity and type of shadows can be made while considering color consistency. The method of calculating the target value enhances the ability to identify shadowed areas, especially in complex urban environments, effectively reducing the impact of changes in lighting or other interference factors on the detection results, thereby improving the anti-interference capability and detection accuracy of the monitoring system.

[0015] Furthermore, the feature vector is: , For the first A feature vector with a given orientation. For the first A set x-coordinate of the orientation. For the first A set ordinate for a given orientation. For the first An azimuth angle for a given orientation. For the first The possible shadows in a given location, For the first The color uniformity of each set direction.

[0016] The aforementioned technical solution significantly improves the accurate extraction and recognition of shadow boundaries by constructing feature vectors that organically combine the color characteristics and spatial orientation information of shadows. The feature vectors not only consider the color possibilities and uniformity of shadows but also incorporate spatial location information such as azimuth angles, enabling the algorithm to more comprehensively describe the characteristics of shadow areas. This comprehensive feature description enhances the system's sensitivity to shadow morphology and location, helps to accurately distinguish shadows from the background in complex environments, improves adaptability to dynamic changes and robustness to various interference factors, thereby improving the overall recognition accuracy and reliability of the monitoring system.

[0017] Furthermore, the behavior analysis algorithm includes optical flow or trajectory tracking algorithms.

[0018] The aforementioned technical solutions, by introducing optical flow or trajectory tracking algorithms for behavior analysis, can effectively track the motion trajectory and changing trends of targets in dynamic scenes. Optical flow captures the motion information of pixels in an image, enabling real-time estimation of the target's speed and direction, thereby identifying the target's movement behavior. Trajectory tracking algorithms further improve the accuracy of target behavior recognition by analyzing the target's positional changes at multiple time points. The application of these algorithms significantly enhances the system's responsiveness to dynamic changes in targets, enabling more stable and accurate identification and prediction of target behavior in complex environments, thus improving the intelligence level of image processing and target analysis.

[0019] Furthermore, CCD or CMOS cameras are used to acquire images during the logistics and transportation process.

[0020] Furthermore, the foreground of the image is extracted using an erosion operation.

[0021] In a second aspect, the present invention provides a logistics violation identification device based on surveillance video, including a memory and a processor. The memory stores computer program instructions, which, when executed by the processor, implement the logistics violation identification method based on surveillance video as described above.

[0022] The beneficial effects of this invention are as follows: This invention achieves intelligent monitoring of the logistics transportation process through image processing technology. It can effectively identify shadowed areas on roads and determine logistics conditions by analyzing the color characteristics and stability of the shadows. This method, through multi-faceted color uniformity analysis and shadow probability assessment, can accurately extract and track abnormal changes on the road, improving the reliability and real-time performance of monitoring. Furthermore, through region growing algorithms and behavioral analysis, it can further eliminate false positives and noise, improving the accuracy of monitoring. This technical solution has strong adaptability in complex logistics transportation environments and can respond quickly in dynamically changing logistics transportation scenarios, providing strong technical support for logistics transportation. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a method for identifying logistics violations based on surveillance video according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the structural block diagram of a logistics violation identification device based on surveillance video according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating a first objective according to an embodiment of the present invention. Detailed Implementation

[0024] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] An example of a method for identifying logistics violations based on surveillance video.

[0026] like Figure 1 The flowchart shown below illustrates a logistics violation identification method based on surveillance video according to an embodiment of the present invention, which includes the following steps: S1: The first objective is obtained by performing an erosion operation on the foreground connected region in the logistics transportation process image, and the second objective is obtained by a subtraction operation.

[0027] like Figure 3 The diagram shows a first objective of an embodiment of the present invention.

[0028] In one embodiment, high-quality images of the logistics transportation process are first captured using a CCD or CMOS camera to obtain high-definition image data.

[0029] To remove noise from images caused by variations in shooting conditions or the surrounding environment, the image is first filtered. Common filtering methods include Gaussian filtering, mean filtering, and morphological filtering. The main goal of filtering is to smooth the image, removing unnecessary noise and details, thereby making subsequent object detection and feature extraction more accurate. For example, Gaussian filtering can effectively smooth the image, reducing brightness variations caused by different lighting conditions or shooting angles, thus reducing interference factors such as light spots and shadows in the image. In addition, the filtering process can also help enhance the edge details of the image, providing a clearer outline for the extraction of foreground objects.

[0030] After filtering, the foreground region of the image typically contains target objects, such as moving vehicles, pedestrians, or road markings. To further extract these foreground targets and remove background noise, the next step is to perform an erosion operation on the foreground connected regions in the image. Erosion is a common morphological processing method that scans the image using structuring elements, thereby reducing the boundaries of the foreground region and removing some isolated noise points. Erosion effectively reduces small noise points in the image and highlights the main target area. After the erosion operation, a first target is obtained. A second target is obtained by subtracting the first target from the foreground connected regions in the image. By subtracting the eroded first target from the foreground connected regions in the image, the subtraction operation preserves the uneroded details, thus extracting a more precise target region.

[0031] S2: Using the geometric center of the second target as the center, obtain points in different directions on the unit circle, and calculate the color uniformity and shadow probability in different directions.

[0032] In one embodiment, the shadow probability is: In the formula, For the first The possible shadows in a given location, This represents the distance from the coordinates corresponding to the first color in the RGB color space to the point (1,1,1). This represents the distance from the coordinates corresponding to the first color in the RGB color space to the point (0,0,0). It is an exponential function with the natural constant e as its base. The maximum value among the RGB three channels of the first color. The first color is the minimum value among the three RGB channels of the first color; the first color is the RGB value that appears most frequently, and the RGB value is the average value of the three RGB channels of each intersection point. The intersection point is the point where a radial line drawn from the geometric center of the second target to a set direction intersects with the second target.

[0033] In another embodiment, the first color is: a radial line is drawn with the geometric center of the second target as the center and directed to a set direction. The intersection of the radial line and the second target is the target pixel. The boundary color corresponding to the set direction is constructed based on the mean of the RGB three channels of the target pixel. The points corresponding to the boundary colors of each set direction in the RGB color space are clustered to obtain multiple clusters. The boundary color corresponding to the largest cluster is taken as the first color of the set direction.

[0034] By radiating a line from the geometric center of the second target to a predetermined location, and calculating the boundary color using the RGB three-channel mean of the target pixels, and then obtaining the most representative boundary color through cluster analysis, the accuracy and stability of boundary recognition can be effectively improved. By clustering boundary colors from multiple predetermined locations, the scheme can automatically identify and extract the most representative color features, avoiding the influence of manual intervention and environmental changes on color determination.

[0035] By calculating the shadow probability at each set location and combining the geometric distance of the first color in RGB space with the extreme differences of the color channels, the stability and consistency of the target color can be quantified. This method effectively reflects the degree of color variation and the uniformity of the boundary region, thus providing a quantitative stability measure for targets at different locations in an image. This helps to stably identify and track targets in complex environments, especially when there are changes in illumination or background interference, ensuring accurate extraction of target boundaries and reliable determination of target status.

[0036] Color uniformity is: , For the first Color consistency for each set direction The number of boundary colors in the largest cluster. The number of connected components in the foreground of the image. The standard deviation of the distance from the point corresponding to each boundary color in the RGB color space to the center of the largest cluster is given. The standard deviation of the distance from the point corresponding to the boundary color of the largest cluster in the RGB color space to its cluster center. The th in the largest cluster The Euclidean distance from the point corresponding to the boundary color in the RGB color space to the center of its cluster.

[0037] By combining the distribution characteristics of intra-cluster boundary colors and their distance from the cluster center, the above method can quantify the stability of color consistency. This method accurately describes the concentration and consistency of colors by considering the number of maximum intra-cluster boundary colors, the standard deviation of the distance from each boundary color to the cluster center, and the number of connected components in the foreground region. In this way, the system can effectively evaluate the color consistency of different target regions in image analysis, providing a more accurate basis for subsequent target recognition and region segmentation. Especially in dynamic or changing environments, it improves the robustness and accuracy of color recognition, helping to reduce misjudgments caused by color fluctuations or interference.

[0038] S3: Construct feature vectors in each direction and map them onto the boundary of the first target, extracting the longest continuous boundary as the shadow boundary.

[0039] In one embodiment, construct the first Feature vectors of a given orientation , , For the first A target value with a set orientation. For the first A set x-coordinate of the orientation. For the first A set ordinate for a given orientation. For the first An azimuth angle for a given orientation. For the first The possible shadows in a given location, For the first The color uniformity of each set direction.

[0040] The feature vectors in each direction are mapped to points in three-dimensional space and clustered. The average value of the target value in each cluster is calculated. The point corresponding to the feature vector in the cluster with the maximum average value is mapped to the boundary of the first target. The boundary with the longest continuity is extracted as the shadow boundary. The midpoint of the shadow boundary is used as the seed point for region growth to obtain the initial shadow.

[0041] By constructing and clustering feature vectors for each designated location, and combining color shadow probability, color uniformity, and directional coordinates, the color and spatial features of the target can be effectively mapped in three-dimensional space, thereby identifying and extracting the most representative shadow boundaries. This method calculates the average feature value of each cluster through cluster analysis and maps the feature vector corresponding to the maximum average value to the target boundary, accurately locating the shadow region. Then, a region growing algorithm is used to extract the longest continuous shadow boundary from the boundary. This technique not only improves the accuracy of shadow boundary extraction and reduces interference from errors and noise, but also enhances the adaptability to changes in the target region, effectively improving the robustness of image processing in complex environments, especially in the accurate identification and analysis of shadow regions in dynamic scenes.

[0042] S4: If the area ratio of the initial shadow to the shadow boundary is less than a set threshold, the initial shadow is removed, a second shadow is obtained, and behavior recognition is performed.

[0043] The threshold value set above can be 0.3, but it can also be determined according to the actual situation.

[0044] In one embodiment, the midpoint of each shadow boundary is used as a seed point for region growth to obtain each initial shadow; the region growth is a prior art and will not be described in detail in this solution.

[0045] First, an initial shadow is extracted and its area is compared with that of the first target to determine its validity. If the ratio of the initial shadow's area to the first target's area is less than a set threshold, the initial shadow is considered invalid due to noise or other interference factors and is therefore discarded, resulting in a second shadow. After discarding invalid shadows, the behavior analysis stage begins, where optical flow or trajectory tracking algorithms are used for dynamic behavior recognition. Optical flow analyzes pixel motion information between consecutive frames to capture the target's trajectory and speed changes in the image, effectively identifying the moving target's behavior patterns. Trajectory tracking algorithms track the target's position changes across multiple image frames, stably capturing the target's movement path and further enhancing the system's adaptability to complex dynamic scenes.

[0046] Taking the trajectory tracking algorithm as an example: The trajectory tracking algorithm specifically involves: tracking the trajectory of the dynamic object corresponding to the second shadow to obtain its spatial position data at multiple time points; based on the trajectory data, calculating the motion characteristics of the dynamic object, including but not limited to the direction of movement, speed, acceleration, etc., and determining whether the object is in a normal movement state; if the object's motion trajectory conforms to a preset abnormal behavior pattern, triggering an alarm mechanism, issuing a warning signal, and indicating that there may be danger or violation; based on the motion trajectory of the dynamic object, predicting its future motion behavior, and if it is predicted that the object may have abnormal behavior, issuing a warning in advance to prevent potential dangers.

[0047] This invention effectively extracts and analyzes shadow areas and anomalies in logistics transportation by performing foreground connected component erosion, subtraction, and geometric analysis on images. By calculating color uniformity and the probability of color shadows, combined with a region growing algorithm, abnormal shadows can be accurately identified and separated, improving the stability and accuracy of images in logistics transportation under complex lighting conditions. Furthermore, the introduction of behavioral analysis algorithms further enhances the ability to identify potential logistics events, strengthening intelligence and responsiveness. The implementation of this method helps optimize real-time monitoring of logistics transportation, improving safety and maintenance efficiency during the process.

[0048] Example of a logistics violation identification device based on surveillance video: like Figure 2 As shown in the figure, the structural block diagram of the logistics violation identification device based on surveillance video according to an embodiment of the present invention includes a processor and a memory.

[0049] This invention also provides a device for identifying logistics violations based on surveillance video. For example... Figure 2 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the above-described method for identifying logistics violations based on surveillance video according to the present invention.

[0050] The logistics violation identification device based on surveillance video also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0051] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions stored or otherwise maintained by such a computer-readable medium.

[0052] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0053] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for identifying logistics violations based on surveillance video, characterized in that, include: Images of the logistics transportation process are acquired, and multiple shadow boundaries are extracted from the images; the midpoint of each shadow boundary is used as a seed point for region growing to obtain each initial shadow. If the area ratio of the initial shadow to the corresponding shadow boundary is less than a set threshold, the initial shadow is removed from the shadow boundary to obtain the second shadow, and the behavior recognition of the second shadow is performed using a behavior analysis algorithm. The process of extracting the shadow boundary is as follows: perform erosion operation on each connected component in the foreground of the image to obtain the first target, and perform subtraction operation on each connected component and the first target to obtain the second target; Using the geometric center of the second target as the center, points in different directions on the unit circle are obtained. The horizontal and vertical coordinates of the points and the target value are used to construct a feature vector. The target value is inversely correlated with color uniformity and positively correlated with shadow probability. The color uniformity represents the stability of the direction, and the shadow probability represents the abnormality of the direction. The feature vectors of each point are mapped to points in three-dimensional space and clustered. The mean of the target value in each cluster is calculated. The point corresponding to the feature vector in the cluster with the maximum mean is mapped to the boundary of the first target. The boundary with the longest continuity is obtained as the shadow boundary. The possibility of the shadow is: In the formula, For the first The possible shadows in a given location, This represents the distance from the coordinates corresponding to the first color in the RGB color space to the point (1,1,1). This represents the distance from the coordinates corresponding to the first color in the RGB color space to the point (0,0,0). It is an exponential function with the natural constant e as its base. The maximum value among the RGB three channels of the first color. It is the minimum value among the three RGB channels of the first color; The first color is the RGB value that appears most frequently. The RGB value is the average of the three RGB channels of each intersection point. The intersection point is the point where a radial line drawn from the geometric center of the second target to a set direction intersects the second target. Color uniformity is: , For the first Color consistency for each set direction The number of boundary colors in the largest cluster. The number of connected components in the foreground of the image. The standard deviation of the distance from the point corresponding to each boundary color in the RGB color space to the center of the largest cluster is given. The standard deviation of the distance from the point corresponding to the boundary color of the largest cluster in the RGB color space to its cluster center. The th in the largest cluster The Euclidean distance from the point corresponding to the boundary color in the RGB color space to the center of its cluster.

2. The method for identifying logistics violations based on surveillance video according to claim 1, characterized in that, The first color is: Radial lines are drawn from the geometric center of the second target to a set direction. The intersection of the radial lines with the second target is the target pixel. The boundary color corresponding to the set direction is constructed based on the mean of the RGB three channels of the target pixel. Cluster the points corresponding to the boundary colors of each set direction in the RGB color space to obtain multiple clusters, and take the boundary color corresponding to the largest cluster as the first color of the set direction.

3. The method for identifying logistics violations based on surveillance video according to claim 1, characterized in that, The target value is: , For the first A target value with a set orientation. For the first The possible shadows in a given location, For the first The color uniformity of each set direction.

4. The method for identifying logistics violations based on surveillance video according to claim 1, characterized in that, The feature vector is: , For the first A feature vector with a given orientation. For the first A set x-coordinate of the orientation. For the first A set ordinate for a given orientation. For the first An azimuth angle for a given orientation. For the first The possible shadows in a given location, For the first The color uniformity of each set direction.

5. The method for identifying logistics violations based on surveillance video according to claim 1, characterized in that, The behavior analysis algorithm includes optical flow or trajectory tracking algorithms.

6. The method for identifying logistics violations based on surveillance video according to claim 1, characterized in that, Use CCD or CMOS cameras to acquire images during the logistics and transportation process.

7. The method for identifying logistics violations based on surveillance video according to claim 1, characterized in that, The foreground of the image is extracted using an erosion operation.

8. A logistics violation identification device based on surveillance video, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the logistics violation identification method based on surveillance video as described in any one of claims 1 to 7 is implemented.

Images