Abnormal resident target detection method and system on highway based on combined background

Abstract

The invention discloses a method and system for detecting abnormal resident targets on a highway based on a combined background. The method comprises: collecting monitoring images of a highway, extracting and caching an original background image; obtaining a T0 frame monitoring image, extracting and caching a current background image, superimposing the original background image with the current background image, updating an image of a suspected abnormal resident target area in the current background image with the original background image, and caching the image as a combined background image; obtaining a Tn frame monitoring image, extracting and caching the current background image, performing frame difference processing on the current background image and the latest combined background image, and updating a list of suspected abnormal resident targets according to the difference image; repeating the above steps until any object in the list of suspected abnormal resident targets meets a reporting condition, using a trained target detection model to perform target recognition on the current frame monitoring image, adding attribute labels to the suspected abnormal resident targets, updating the list of suspected abnormal resident targets and reporting the target.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method and system for detecting abnormal resident targets on highways based on combined backgrounds. Background Art

[0002] With the continuous increase in traffic volume on highways, abnormal residence has become one of the important factors affecting road safety. Abnormal residence refers to the situation where vehicles, people or other objects stay on the highway for more than a certain threshold, which may pose a serious threat to the safety of highway traffic, including illegally parked vehicles or people, spilled objects, obstacles, etc. Traditional detection of abnormal residence targets on highways mainly relies on manual inspections or video surveillance with fixed cameras. These methods are not only inefficient, but also easily interfered by complex environmental factors, resulting in inaccurate detection or missed detection, and unable to provide timely warnings and responses.

[0003] Existing automated video surveillance technologies usually use background modeling methods to detect abnormal targets. These methods mainly include background modeling techniques such as single Gaussian models and mixed Gaussian models. By modeling stationary or slow-moving targets in highway scenes, resident objects can be effectively detected. However, in complex dynamic environments such as highways, due to factors such as fast-moving vehicles, lighting changes, and camera jitter, background modeling technology often faces the following problems when dealing with abnormal resident targets:

[0004] 1. High false alarm and missed alarm rates: When facing the dynamic background on the highway, the existing background modeling technology is prone to misreporting targets that stay for a short period of time as abnormal targets, or missing the actual abnormal resident targets. Especially when facing similar environmental changes (such as sudden changes in light), the system's detection accuracy is difficult to guarantee.

[0005] 2. Limitations of deep learning detection technology: Although deep learning technology (such as the YOLO target detection algorithm) has made significant progress in the field of target detection in recent years, the uncertainty of abnormal resident targets in highway scenarios poses a challenge to deep learning algorithms. Existing deep learning methods usually rely on a large amount of pre-labeled target data, but abnormal resident targets on highways are often random and difficult to predict, resulting in frequent false positives and false negatives. At the same time, such algorithms often have high requirements for GPU resources and are difficult to be widely used when computing resources are limited. Summary of the invention

[0006] In order to solve the technical problems existing in the above-mentioned prior art, the present invention aims to provide a method and system for detecting abnormal resident targets on highways based on a combination of combined background images and deep learning detection technology. By dynamically combining background images of different time periods, a more accurate combined background is generated, which can effectively cope with the complex dynamic background on the highway. At the same time, the detected targets are further classified and labeled in combination with the deep learning model, which improves the detection accuracy and real-time performance of the system, and can effectively operate when CPU resources are sufficient and GPU resources are limited, making up for the shortcomings of the existing technology and significantly improving the accuracy and efficiency of abnormal resident target detection on highways.

[0007] In order to achieve the above-mentioned invention object, the technical solution provided by the present invention includes:

[0008] The method for detecting abnormal resident targets on highways based on combined backgrounds comprises the following steps:

[0009] S1, collecting monitoring images of the target area of ​​the expressway, extracting and caching the original background image;

[0010] S2, acquiring the T0 frame monitoring image, extracting and caching the current background image, superimposing the original background image with the current background image, updating the image of the suspected abnormality resident target area in the current background image with the original background image, and caching it as a combined background image;

[0011] S3, obtaining Tn frames of monitoring images, extracting and caching the current background image, performing frame difference processing between the current background image and the latest combined background image, and updating the suspected abnormal resident target list according to the difference image, where n=1,2,3,…; when the suspected abnormal resident target list changes or reaches a preset update cycle, caching a new combined background image;

[0012] S4. Repeat step S3 until any object in the suspected abnormal resident target list meets the reporting conditions, use the trained target detection model to perform target recognition on the current frame monitoring image, add attribute labels to the suspected abnormal resident targets, update the suspected abnormal resident target list and report it; the reporting conditions include a preset reporting period.

[0013] Preferably, the suspected abnormal resident target list includes: suspected abnormal resident target ID, suspected abnormal resident target image, suspected abnormal resident target appearance time, duration, disappearance time and continuous disappearance time.

[0014] Preferably, the method for caching the new combined background image in step S3 includes:

[0015] Acquire the current frame monitoring image, extract and cache the current background image, superimpose the current background image with the latest combined background image, update the image of the suspected abnormal resident target area in the current background image with the combined background image, and cache it as a new combined background image.

[0016] Preferably, the updating of the suspected abnormal resident target list includes:

[0017] When the circumscribed rectangle of the new difference image contour area meets the preset width and / or height threshold, the suspected abnormal resident target is updated with the new difference image contour area;

[0018] When the new difference image contour overlaps with the current difference image contour, the suspected abnormal resident target is updated with the union of the new difference image contour area and the current difference image contour area;

[0019] When the new difference image contour includes the current difference image contour, updating the suspected abnormal resident target with the new difference image contour area;

[0020] When the new difference image contour does not overlap with the current difference image contour, and this state is maintained for a preset time, the suspected abnormal resident target is updated with the new difference image contour area.

[0021] Preferably, the reporting conditions also include:

[0022] The duration of the suspected abnormal target residence meets a preset time threshold;

[0023] The attribute label of the suspected abnormal resident target is unknown.

[0024] The present invention also discloses a highway abnormal resident target detection system based on combined background, comprising:

[0025] An image acquisition module is used to acquire monitoring images of a target area of ​​a highway;

[0026] A cache module, used to cache a list of suspected abnormal resident targets and background images;

[0027] A background update module, connected to the image acquisition module and the cache module respectively, for extracting and caching the original background image according to the monitoring image; for acquiring the T0 frame monitoring image, extracting and caching the current background image, superimposing the original background image with the current background image, and updating the image of the suspected abnormal resident target area in the current background image with the original background image, and caching it as a combined background image; and for caching a new combined background image when the suspected abnormal resident target list changes or reaches a preset update cycle;

[0028] A suspected abnormal resident target acquisition module is connected to the image acquisition module, the cache module and the background update module respectively, and is used to acquire Tn frame monitoring images, extract and cache the current background image, perform frame difference processing on the current background image and the latest combined background image, and update the suspected abnormal resident target list according to the difference image, wherein n=1,2,3,…;

[0029] A target detection module, which has a built-in trained target detection model and is connected to the image acquisition module and the cache module respectively, and is used to identify the target in the current frame monitoring image when any object in the suspected abnormal resident target list meets the reporting condition, add an attribute label to the suspected abnormal resident target, and update the suspected abnormal resident target list;

[0030] A reporting module is connected to the cache module and is used to report the suspected abnormal resident targets in the suspected abnormal resident target list that meet the reporting conditions; the reporting conditions include a preset reporting period.

[0031] Preferably, the method for the background update module to cache the new combined background image includes:

[0032] Acquire the current frame monitoring image, extract and cache the current background image, superimpose the current background image with the latest combined background image, update the image of the suspected abnormal resident target area in the current background image with the combined background image, and cache it as a new combined background image.

[0033] Preferably, the list of suspected abnormal resident targets includes: suspected abnormal resident target ID, suspected abnormal resident target image, suspected abnormal resident target appearance time, duration, disappearance time and continuous disappearance time;

[0034] The update of the suspected abnormal resident target list includes:

[0035] When the circumscribed rectangle of the new difference image contour area meets the preset width and / or height threshold, the suspected abnormal resident target is updated with the new difference image contour area;

[0036] When the new difference image contour overlaps with the current difference image contour, the suspected abnormal resident target is updated with the union of the new difference image contour area and the current difference image contour area;

[0037] When the new difference image contour includes the current difference image contour, updating the suspected abnormal resident target with the new difference image contour area;

[0038] When the new difference image contour does not overlap with the current difference image contour, and this state is maintained for a preset time, the suspected abnormal resident target is updated with the new difference image contour area.

[0039] Beneficial Effects

[0040] 1. Improve detection accuracy: The combined background technology proposed in the present invention can effectively cope with the complex dynamic background of highways, such as fast passing of vehicles, lighting changes, camera shaking, etc., reducing the false positives and missed negatives in traditional background modeling methods, and can more accurately identify abnormal resident targets, especially those that stay for a long time or disappear for a short time and then reappear.

[0041] 2. Enhance the real-time performance and response speed of the system: Through the combination of dynamic background update and deep learning detection technology, abnormal targets can be quickly extracted and the target list can be updated in time to achieve efficient real-time monitoring and abnormal alarm, meeting the requirements for timely response to events in highway environments.

[0042] 3. Reduce computing resource consumption: The present invention uses a combination of background technologies to greatly reduce dependence on deep learning detection and establishes the calling frequency of the target detection model, thereby reducing the system's demand for computing resources, significantly reducing system costs, and improving the system's scalability and adaptability.

[0043] 4. Improve the intelligence and automation level of the system: The present invention combines deep learning target detection technology, which can automatically add attribute labels to suspected abnormal resident targets and automatically determine whether to report based on preset conditions, greatly improving the intelligence level of the system and reducing dependence on manual intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A schematic flow chart of a method for detecting abnormal resident targets on highways based on combined backgrounds provided in a preferred embodiment of the present invention;

[0045] Figure 2 A schematic diagram of the image change process of the background image extracted by using the Gaussian mixture model provided by the prior art during the period between the appearance of the scattered objects and the disappearance of the scattered objects;

[0046] Figure 3 A schematic diagram of a combined background generation process provided in another preferred embodiment of the present invention;

[0047] Figure 4 A schematic diagram of the process and result of performing frame difference processing on the current frame monitoring image and the combined background image to obtain a difference image provided in another preferred embodiment of the present invention;

[0048] Figure 5The schematic diagram of the structure of a highway abnormal resident target detection system based on combined background is provided in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described below in conjunction with the accompanying drawings. In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside" and the like indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention.

[0050] Embodiment 1

[0051] like Figure 1 As shown, this embodiment provides a method for detecting abnormal resident targets on a highway based on a combined background, comprising the steps of:

[0052] S1. Collect monitoring images of the target area of ​​the highway, extract and cache the original background image.

[0053] The target area refers to a specific area within the monitoring range of the expressway where abnormal resident target detection is required, including but not limited to the main roads, emergency parking strips, ramps and connecting sections, bridges and tunnels of the expressway. The abnormal resident target refers to an object that appears in a specific target area (such as the main road, emergency parking strip, ramp, etc.) in the expressway scene and stays for a time exceeding the normal threshold, including but not limited to vehicles that stay for a long time, spilled objects or obstacles, pedestrians or animals, and other stationary or slow-moving objects. These targets should not appear in specific areas of the expressway for a long time, and their stay is usually unsafe or illegal, and often poses a threat to traffic safety, such as obstructing traffic flow or causing traffic accidents.

[0054] The background image refers to a static scene image extracted from a highway surveillance video, which is a reference image commonly used in the prior art to identify and separate abnormal resident targets. The extraction method can be based on a variety of background modeling techniques, such as a mixed Gaussian model (GMM), using statistical information such as the probability density of a large number of sample values ​​of pixels over a long period of time (such as the number of patterns, the mean and standard deviation of each pattern) to represent the background, and then using statistical differences (such as the 3σ principle) to judge the target pixel, which can model complex dynamic backgrounds. It should be understood that the extraction of static background images is not the focus of the present invention, so those skilled in the art can select a suitable method in the prior art to implement it, and the present invention does not make further limitations.

[0055] The original background image is a background image initially collected in a monitoring scene and is used as a reference for subsequent image processing. The original background image does not contain any highway area image of any active targets (such as vehicles or people).

[0056] S2. Acquire the T0 frame monitoring image, extract and cache the current background image, superimpose the original background image with the current background image, update the image of the suspected abnormal resident target area in the current background image with the original background image, and cache it as a combined background image.

[0057] The T0 frame monitoring image refers to the monitoring image at the starting time after the original background image is completed, and it is used as the first monitoring image frame collected when the system is started or reinitialized.

[0058] When using background modeling technology to detect spilled objects, the existing technology can ignore fast-moving vehicles and other targets, and only use logical means to process the images extracted from the stationary targets or slow-moving targets within a period of time on the highway. Figure 2 As shown in the figure, the background image extracted by the Gaussian mixture model changes between the time when the scattered objects appear and the time when the scattered objects disappear. Although this technology can cope with the changes in dynamic background to a certain extent, it also has certain limitations, especially in dealing with fast-moving vehicles on highways, lighting changes, camera shaking and other factors, resulting in high false alarm rates or missed alarm rates.

[0059] To this end, the present invention considers using the original background image obtained by traditional background modeling as a basis, and then superimposing the background of the image with the suspected abnormal resident target and the original background image to generate a combined background image by superimposing the two, such as Figure 3 The combined background image has at least the following advantages:

[0060] 1. By dynamically updating the background information of different time periods, it is possible to better filter out dynamic targets that appear briefly (such as vehicles passing by quickly), and only retain abnormal targets that truly reside in the scene to reflect their continuous state, thereby improving the accuracy of detection and reducing the occurrence of false alarms.

[0061] 2. It can be adjusted more flexibly according to environmental changes and dynamically update the background to ensure that the system can work stably under various complex conditions. Compared with traditional background images, combined background images provide a more accurate background benchmark, which can provide a more effective detection basis for subsequent deep learning models, further improving the accuracy and efficiency of the system.

[0062] S3. Obtain Tn frame monitoring images, extract and cache the current background image, perform frame difference processing on the current background image and the latest combined background image, and update the list of suspected abnormal resident targets according to the difference image, where n=1, 2, 3, ...; when the list of suspected abnormal resident targets changes or reaches a preset update cycle, cache a new combined background image.

[0063] The Tn frame monitoring images refer to frame-by-frame monitoring images after the first combined background image cache is completed, and n represents a dynamically increasing frame number.

[0064] The frame difference processing refers to detecting and extracting the changed parts of the image by comparing the differences between two or more consecutive video frames. Figure 4 As shown, a difference image is obtained by performing frame difference processing on the background image of the current frame monitoring image and the combined background image, and then a binarization analysis is performed on the frame difference image, and then a contour search is performed in the binarized image to obtain the contour area of ​​the difference image. If there are multiple contour areas in the difference image, a contour object list needs to be established. Since the combined background image in the present invention has excluded dynamic change factors (such as fast-moving vehicles, light changes, etc.) as much as possible, and further, the current background image and the combined background image are used for frame difference processing, and the difference relationship between the dynamic target and the static background reflected by the traditional difference between the current frame image and the background image is converted into the difference relationship between the static background and the static background. This technical solution has the following technical advantages:

[0065] 1. It can effectively suppress short-term or small dynamic changes (such as light fluctuations, temporary occlusion, etc.), because these interferences will be smoothed out in the superimposed combined background image, thereby reducing false alarms caused by dynamic noise;

[0066] 2. Since the combined background image itself is a successive superposition and update of multiple static background images, it can reflect the long-term resident objects in the environment. For abnormal resident targets (such as vehicles that stay on the highway for a long time), this static background comparison can more clearly capture the persistent difference between it and the background, rather than short-term dynamic changes, and improve the accuracy of detection.

[0067] 3. In accordance with the update conditions of the combined background image, the combined background image can be updated when the list of suspected abnormal resident targets changes or reaches the preset update cycle, instead of updating every frame. This helps to reduce resource consumption caused by frequent updates, making the system more efficient, and is particularly suitable for long-term large-scale monitoring.

[0068] 4. It can better adapt to long-term changes in dynamic environments (such as changes in light and climate), reflect the trend of environmental changes, and enable the detection system to have better adaptability in different scenarios.

[0069] It should be noted that the cache of the combined background image should reflect the characteristics of the current monitoring image. Therefore, in addition to caching a combined background image when the initial run or the current cached background image is empty, in some preferred embodiments, specific cache update conditions are also given, as follows:

[0070] 1. When the list of suspected abnormal resident targets changes, the combined background image needs to be updated. Specific changes may include: an increase or decrease in the number of targets, a change in the location of the target, a change in the size of the target, etc.

[0071] 2. When the preset update period is reached, even if there is no obvious target change in the monitoring area, the combined background image will be updated. This update mechanism is mainly to cope with some long-term changes in the environment, such as slow changes in light conditions, slight camera shaking, etc. Although these changes do not involve obvious movement or change of the target, they may also affect the stability of the background image.

[0072] The specific updating method is similar to the above steps, obtaining the current frame monitoring image, extracting and caching the current background image, superimposing the current background image with the latest combined background image, and using the combined background image to update the image of the suspected abnormal target area in the current background image, and caching it as an updated combined background image.

[0073] In addition to being updated when a new suspected abnormal resident target appears or a new attribute tag is added, the suspected abnormal resident target list also needs to set update conditions according to the resident characteristics of the suspected abnormal resident target. In some preferred embodiments, the update of the suspected abnormal resident target list includes:

[0074] 1. When the bounding rectangle of the new difference image contour area meets the preset width and / or height thresholds, it means that if the contour size of an object exceeds a specific threshold, it will be considered as an abnormal resident target, so the suspected abnormal resident target is updated with the new difference image contour area.

[0075] 2. When the new difference image contour overlaps with the current difference image contour, they are considered to be the same target, but the volume may increase or the position may move. Therefore, in order to ensure that the morphological changes of the suspected abnormal resident target can be tracked, the union of the new difference image contour area and the current difference image contour area is used to update the suspected abnormal resident target.

[0076] 3. When the new difference image contour contains the current difference image contour, it is considered that the new object is a replacement of the old object, or is a more detailed target, so the suspected abnormal resident target is updated with the new difference image contour area.

[0077] 4. When the new difference image contour does not overlap with the current difference image contour and this state is maintained for a preset time, it is considered that a new suspected abnormal resident target has appeared, so the suspected abnormal resident target is updated with the new difference image contour area.

[0078] The above update conditions cover multiple judgment criteria related to the size of the detection target, overlapping relationship and time threshold, ensuring that the system can update the target information in real time and accurately and track abnormal resident objects on the highway. The setting of these conditions can effectively reduce false positives and false negatives and improve the accuracy and real-time performance of the detection system.

[0079] It should be noted that the suspected abnormal resident target of the present invention refers to the target set of all possible abnormal resident behaviors detected by the system and all the information contained in these targets. The present invention uses a suspected abnormal resident target list to record its relevant information. The suspected abnormal resident target is an individual in the suspected abnormal resident target, which refers to one or more specific object entities in a specific detection area. The object may be a vehicle, a cargo, a pedestrian, etc., which is identified by the system as an object that may have abnormal resident behavior. Each suspected abnormal resident target will have its own attributes, such as ID, location, residence time, etc. Through this information, the system can track and judge each specific object. In short, the suspected abnormal resident target is a broader set, which contains multiple suspected abnormal resident targets, and the latter is a specific individual in the former, representing an actual entity object. In some preferred embodiments, a preferred example of the suspected abnormal resident target list is given, specifically including: suspected abnormal resident target ID, suspected abnormal resident target, suspected abnormal resident target appearance time, duration, disappearance time and continuous disappearance time. It should be understood that the suspected abnormal resident target includes the video image itself where it is located, as well as information such as its outline and position in the image.

[0080] S4. Repeat step S3 until any object in the suspected abnormal resident target list meets the reporting conditions, use the trained target detection model to perform target recognition on the current frame monitoring image, add attribute labels to the suspected abnormal resident targets, update the suspected abnormal resident target list and report it; the reporting conditions include a preset reporting period.

[0081] The target detection model refers to a machine learning or deep learning model used to identify and classify target objects (such as vehicles, objects, pedestrians, etc.) that appear in images or videos. This type of model can automatically identify specific targets in the image by analyzing the feature information in the image, and add attribute labels to these targets, such as the type, location, size and other information of the target. Commonly used models include but are not limited to YOLO series models, Faster R-CNN, etc. The present invention does not intend to limit the specific target detection model type and its training method, as long as it can achieve target detection.

[0082] It should be noted that the technical considerations for the present invention to perform target recognition before reporting rather than from the beginning are as follows: when performing target detection, the number of images detected per unit time is proportional to the computing power consumption. The existing technology performs target detection on each monitoring image, resulting in excessively high computing costs. However, the present invention performs target detection on a single-frame monitoring image after determining the presence of an abnormal resident target, thereby reducing the use of the target recognition model, thereby reducing the amount of calculation and improving the operating efficiency of the system.

[0083] Specifically, the reported information includes but is not limited to: video images and their timestamps within a certain period of time from the appearance to the disappearance of the suspected abnormal resident target in the target area, event types (this label is related to the attribute label of the suspected abnormal resident target, including but not limited to illegally parked persons or vehicles, spilled objects or intrusive objects, etc.), event status identification (initialization, event start, event continuation, event disappearance), and suspected abnormal resident target information.

[0084] The reporting conditions refer to the fact that in the present invention, when the detected suspected abnormal resident target meets certain specific conditions, the system will report the target information or notify the relevant departments. These reporting conditions are an important basis for the system to determine which abnormal targets need further processing. The setting of reporting conditions can ensure that the system only reports abnormal situations that really need attention, avoid excessive reporting, and reduce the burden on monitoring personnel.

[0085] In addition to the preset reporting period, in some preferred embodiments, the following reporting conditions are also provided, specifically including:

[0086] 1. The duration of the suspected abnormal residence target meets the preset time threshold. If the suspected abnormal residence target resides in the monitoring area for longer than the set time threshold (such as 30 seconds, 1 minute, etc.), the target is considered to have abnormal residence behavior and triggers reporting. For example, if a car is parked in an emergency parking strip or on a highway main road for more than the set time limit, the system will automatically report the abnormal residence situation.

[0087] 2. The attribute label of the suspected abnormal resident target is unknown. At this time, it means that the label of the target cannot be matched with the known target type (such as vehicles, pedestrians, scattered objects, etc.), and the system will mark the object as "unknown". The "unknown" attribute usually means that the type of the target cannot be clearly identified by the system, which may be an abnormal situation. The system will report such targets for further manual verification.

[0088] Embodiment 2

[0089] like Figure 5 As shown, this embodiment provides a highway abnormal resident target detection system based on combined background, including:

[0090] An image acquisition module is used to acquire monitoring images of a target area of ​​a highway;

[0091] A cache module, used to cache a list of suspected abnormal resident targets and background images;

[0092] A background update module, connected to the image acquisition module and the cache module respectively, for extracting and caching the original background image according to the monitoring image; for acquiring the T0 frame monitoring image, extracting and caching the current background image, superimposing the original background image with the current background image, and updating the image of the suspected abnormal resident target area in the current background image with the original background image, and caching it as a combined background image; and for caching a new combined background image when the suspected abnormal resident target list changes or reaches a preset update cycle;

[0093] A suspected abnormal resident target acquisition module is connected to the image acquisition module, the cache module and the background update module respectively, and is used to acquire Tn frame monitoring images, extract and cache the current background image, perform frame difference processing on the current background image and the latest combined background image, and update the suspected abnormal resident target list according to the difference image, wherein n=1,2,3,…;

[0094] A target detection module, which has a built-in trained target detection model and is connected to the image acquisition module and the cache module respectively, and is used to identify the target in the current frame monitoring image when any object in the suspected abnormal resident target list meets the reporting condition, add an attribute label to the suspected abnormal resident target, and update the suspected abnormal resident target list;

[0095] A reporting module is connected to the cache module and is used to report the suspected abnormal resident targets in the suspected abnormal resident target list that meet the reporting conditions; the reporting conditions include a preset reporting period.

[0096] In some preferred embodiments, the method for the background update module to cache the new combined background image includes:

[0097] Acquire the current frame monitoring image, extract and cache the current background image, superimpose the current background image with the latest combined background image, update the image of the suspected abnormal resident target area in the current background image with the combined background image, and cache it as the updated combined background image.

[0098] In some preferred embodiments, the list of suspected abnormal resident targets includes: suspected abnormal resident target ID, suspected abnormal resident target image, suspected abnormal resident target appearance time, duration, disappearance time and continuous disappearance time;

[0099] The update of the suspected abnormal resident target list includes:

[0100] When the circumscribed rectangle of the new difference image contour area meets the preset width and / or height threshold, the suspected abnormal resident target is updated with the new difference image contour area;

[0101] When the new difference image contour overlaps with the current difference image contour, the suspected abnormal resident target is updated with the union of the new difference image contour area and the current difference image contour area;

[0102] When the new difference image contour includes the current difference image contour, updating the suspected abnormal resident target with the new difference image contour area;

[0103] When the new difference image contour does not overlap with the current difference image contour, and this state is maintained for a preset time, the suspected abnormal resident target is updated with the new difference image contour area.

[0104] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for detecting abnormal resident targets on highways based on combined backgrounds, characterized in that: Includes steps: S1, collecting monitoring images of the target area of ​​the expressway, extracting and caching the original background image; S2, acquiring the T0 frame monitoring image, extracting and caching the current background image, superimposing the original background image with the current background image, updating the image of the suspected abnormality resident target area in the current background image with the original background image, and caching it as a combined background image; S3, obtaining Tn frames of monitoring images, extracting and caching the current background image, performing frame difference processing between the current background image and the latest combined background image, and updating the suspected abnormal resident target list according to the difference image, where n=1,2,3,…; when the suspected abnormal resident target list changes or reaches a preset update cycle, caching a new combined background image; S4, repeating step S3 until any object in the suspected abnormal resident target list meets the reporting condition, using the trained target detection model to perform target recognition on the current frame monitoring image, adding attribute labels to the suspected abnormal resident targets, updating the suspected abnormal resident target list and reporting; the reporting condition includes a preset reporting period; The method for caching the new combined background image in step S3 includes: Acquire the current frame monitoring image, extract and cache the current background image, superimpose the current background image with the latest combined background image, update the image of the suspected abnormal resident target area in the current background image with the combined background image, and cache it as a new combined background image.

2. The method for detecting abnormal resident targets on highways based on combined background according to claim 1, characterized in that: The suspected abnormal resident target list includes: suspected abnormal resident target ID, suspected abnormal resident target image, suspected abnormal resident target appearance time, duration, disappearance time and continuous disappearance time.

3. The method for detecting abnormal resident targets on highways based on combined background according to claim 2, characterized in that: The update of the suspected abnormal resident target list includes: When the circumscribed rectangle of the new difference image contour area meets the preset width and / or height threshold, the suspected abnormal resident target is updated with the new difference image contour area; When the new difference image contour overlaps with the current difference image contour, the suspected abnormal resident target is updated with the union of the new difference image contour area and the current difference image contour area; When the new difference image contour includes the current difference image contour, updating the suspected abnormal resident target with the new difference image contour area; When the new difference image contour does not overlap with the current difference image contour, and this state is maintained for a preset time, the suspected abnormal resident target is updated with the new difference image contour area.

4. The method for detecting abnormal resident targets on highways based on combined background according to claim 2, characterized in that: The reporting conditions also include: The duration of the suspected abnormal target residence meets a preset time threshold; The attribute label of the suspected abnormal resident target is unknown.

5. The highway abnormal resident target detection system based on combined background is characterized by: include: An image acquisition module is used to acquire monitoring images of a target area of ​​a highway; A cache module, used to cache a list of suspected abnormal resident targets and background images; A background update module, connected to the image acquisition module and the cache module respectively, for extracting and caching an original background image according to the monitoring image; It is also used to obtain the T0 frame monitoring image, extract and cache the current background image, superimpose the original background image with the current background image, update the suspected abnormal resident target area image in the current background image with the original background image, and cache it as a combined background image; it is also used to cache a new combined background image when the suspected abnormal resident target list changes or reaches a preset update cycle; A suspected abnormal resident target acquisition module is connected to the image acquisition module, the cache module and the background update module respectively, and is used to acquire Tn frame monitoring images, extract and cache the current background image, perform frame difference processing on the current background image and the latest combined background image, and update the suspected abnormal resident target list according to the difference image, wherein n=1,2,3,…; A target detection module, which has a built-in trained target detection model and is connected to the image acquisition module and the cache module respectively, and is used to identify the target in the current frame monitoring image when any object in the suspected abnormal resident target list meets the reporting condition, add an attribute label to the suspected abnormal resident target, and update the suspected abnormal resident target list; A reporting module, connected to the cache module, used to report suspected abnormal resident targets in the suspected abnormal resident target list that meet the reporting conditions; the reporting conditions include a preset reporting period; The method for the background update module to cache a new combined background image includes: Acquire the current frame monitoring image, extract and cache the current background image, superimpose the current background image with the latest combined background image, update the image of the suspected abnormal resident target area in the current background image with the combined background image, and cache it as a new combined background image.

6. The highway abnormal resident target detection system based on combined background as claimed in claim 5, characterized in that: The suspected abnormal resident target list includes: suspected abnormal resident target ID, suspected abnormal resident target image, suspected abnormal resident target appearance time, duration, disappearance time and continuous disappearance time; The update of the suspected abnormal resident target list includes: When the circumscribed rectangle of the new difference image contour area meets the preset width and / or height threshold, the suspected abnormal resident target is updated with the new difference image contour area; When the new difference image contour overlaps with the current difference image contour, the suspected abnormal resident target is updated with the union of the new difference image contour area and the current difference image contour area; When the new difference image contour includes the current difference image contour, updating the suspected abnormal resident target with the new difference image contour area; When the new difference image contour does not overlap with the current difference image contour, and this state is maintained for a preset time, the suspected abnormal resident target is updated with the new difference image contour area.

Images