Misjudgment prevention system and method for video traffic event detector

Abstract

The invention relates to an anti-misjudgment system and method for a video traffic event detector, and aims to generate enhanced multi-modal data through an advanced image fusion technology so as to effectively cope with interference of complex environments such as strong light, backlight, rain and snow and improve high-quality data in order to reduce misjudgment and improve the judgment precision of complex traffic events. The method comprises the following steps: firstly, carrying out multi-modal data analysis, then realizing accurate foreground target separation, highlighting a motion area, inhibiting background noise and reducing false alarm by adopting a self-adaptive background modeling method on the multi-modal data, and finally, further reducing misjudgment and missed judgment in combination with multi-level step-by-step refined verification so as to realize the purpose of greatly reducing the misjudgment and missed judgment rate of traffic events.

Description

Technical Field

[0001] This invention belongs to the field of image recognition technology and relates to a system and method for preventing false judgments in a video traffic incident detector. Background Technology

[0002] In recent years, with the development of artificial intelligence and drone technology, intelligent transportation systems have also experienced rapid growth. Research on video traffic incident detectors based on computer vision and image processing technologies has attracted widespread attention, and many researchers have made significant progress in this field. However, due to the complexity of the traffic environment, existing technologies still have certain limitations in practical applications.

[0003] Existing video image-based traffic accident detection technologies classify traffic events into 12 categories, including common incidents such as road weather events, traffic accidents, stoppages, congestion, and fireworks. Current regulations require video traffic incident detectors to have a detection rate of no less than 97% and a false alarm rate of no more than 2%, with each video stream detecting no more than one false alarm within 24 hours when the system is in normal detection mode. However, most existing systems have low accuracy in identifying accidents under complex road conditions and severe weather, and are easily affected by factors such as changes in lighting, shadows, and rain / snow, leading to high false alarm and false alarm rates. Furthermore, they struggle to efficiently meet the requirements of regulations, exhibiting technical problems related to high false positive and false negative rates. Summary of the Invention

[0004] To address the problems existing in the above-mentioned traditional methods, this invention proposes a system and method for preventing false judgments in video traffic incident detectors, which can significantly reduce the false judgment and missed judgment rates of traffic incidents.

[0005] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: On the one hand, a false alarm prevention system for video traffic incident detectors is provided, including: The image acquisition module is used to simultaneously acquire visible light video streams and infrared thermal imaging video streams collected by traffic monitoring equipment; The image enhancement module is used to perform image enhancement processing on visible light video streams and infrared thermal imaging video streams, and then use adaptive background modeling to separate foreground targets to generate fused and enhanced image data. The image enhancement processing includes image preprocessing, image registration, feature image decomposition, and fusion reconstruction. The event detection module is used to automatically detect and output the categories of candidate traffic events and related event information based on the input fused image; The false positive prevention module is used to perform multi-evidence chain verification and screening of candidate traffic events from coarse to fine, and output the final traffic event category and false positive confidence. The multi-evidence chain verification and screening includes low-level feature verification, physical law verification, context consistency verification and historical data verification.

[0006] On the other hand, a method for preventing false alarms in a video traffic incident detector is also provided, including the following steps: Simultaneously acquire visible light video streams and infrared thermal imaging video streams collected by traffic monitoring equipment; After image enhancement processing of visible light video stream and infrared thermal imaging video stream, adaptive background modeling is used to separate foreground targets and generate fused and enhanced image data; the image enhancement processing includes image preprocessing, image registration, feature image decomposition and fusion reconstruction. The system automatically detects and outputs the categories of candidate traffic events and related event information based on the input fused image. Candidate traffic events are subjected to progressive verification and screening using a multi-evidence chain from coarse to fine, and the final category of the traffic event and the confidence level for preventing false positives are output. The progressive verification and screening of the multi-evidence chain includes low-level feature verification, physical law verification, contextual consistency verification, and historical data verification.

[0007] One of the above technical solutions has the following advantages and beneficial effects: The aforementioned video traffic incident detector's anti-false alarm system and method generate enhanced multimodal data through advanced image fusion technology to effectively cope with complex environmental interference such as strong light, backlight, and rain and snow. This provides high-quality data to reduce false alarms and improve the accuracy of judging complex traffic incidents. Then, by using an adaptive background modeling method on the multimodal data, accurate foreground target separation is achieved, highlighting the moving area and suppressing background noise, thus reducing false alarms. Finally, multi-level progressively refined verification is combined to further reduce false alarms and missed alarms, thereby achieving the goal of significantly reducing the false alarm and missed alarm rates of traffic incidents. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the 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.

[0009] Figure 1 This is a schematic diagram of the module architecture of a video traffic incident detector's anti-false alarm system in one embodiment; Figure 2 This is a schematic diagram of the application process of the anti-false alarm system for a video traffic incident detector in one embodiment; Figure 3 This is a schematic diagram illustrating the implementation process of the anti-false judgment module in one embodiment; Figure 4 This is a flowchart illustrating a method for preventing false alarms in a video traffic incident detector in one embodiment. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0011] It should be noted that, in this document, the reference to "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The presentation of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will understand that the embodiments described herein can be combined with other embodiments. The term "and / or" as used herein refers to any combination of one or more of the associated listed items, and all possible combinations, including such combinations.

[0012] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0013] Existing methods for detecting road traffic accidents first input video frames, identify vehicles in the video, and calculate vehicle speed, acceleration, and heading angle. Second, these parameters are converted into time-frequency signals, and the resulting time-frequency graph is input into a ResNet50 network for training. Finally, the time-frequency graph features of vehicle speed, acceleration, and heading angle are extracted and compared with those under normal conditions to determine the occurrence of a traffic accident. This method struggles to adapt to scenarios with background interference such as strong light, backlight, rain, snow, and mud stains. The model's detection accuracy needs improvement, and because it relies on a single image source, it suffers from high rates of false positives and false negatives for small or complex objects.

[0014] This invention is based on the concept of multimodal information fusion, which combines infrared light images with visible light images to reduce false alarms and improve the accuracy of judgment on complex traffic events. At the same time, through adaptive background modeling, it can quickly adapt to background interference such as strong light, backlight, rain, snow and mud pollution, reducing false alarms. Furthermore, combined with multi-level progressively refined verification, it ultimately achieves the goal of efficiently reducing false alarms and missed detections.

[0015] In one embodiment, such as Figure 1 As shown, a false positive prevention system for a video traffic incident detector is provided, which may include an image acquisition module 11, an image enhancement module 13, an event detection module 15, and a false positive prevention module 17. The image acquisition module 11 is used to simultaneously acquire visible light video streams and infrared thermal imaging video streams collected by traffic monitoring equipment. The image enhancement module 13 is used to perform image enhancement processing on the visible light video streams and infrared thermal imaging video streams, and then use adaptive background modeling to separate foreground targets, generating fused and enhanced image data; the image enhancement processing includes image preprocessing, image registration, feature image decomposition, and fusion reconstruction. The event detection module 15 is used to automatically detect and output the category of candidate traffic incidents and related event information based on the input fused image. The false positive prevention module 17 is used to perform multi-evidence chain verification and screening of candidate traffic incidents from coarse to fine, and output the final traffic incident category and false positive prevention confidence; the multi-evidence chain verification and screening includes low-level feature verification, physical law verification, contextual consistency verification, and historical data verification.

[0016] Understandable, such as Figure 2 As shown, the system first synchronously acquires visible light video streams and infrared thermal imaging video streams collected by traffic monitoring equipment. After image enhancement processing of these streams by the image enhancement module, enhanced image data is generated to effectively address interference from complex environments such as strong light, backlight, and rain / snow. Based on this, an adaptive background modeling method is used to achieve accurate foreground target separation, highlighting moving areas and suppressing background noise. The image enhancement process of the visible light video stream and infrared thermal imaging video stream can include image preprocessing, image registration, feature image decomposition, and fusion reconstruction. Image preprocessing aims to prepare high-quality and standardized input image data for subsequent image fusion. Image registration ensures that the visible light and infrared images (which are continuously generated with the video stream) synchronously acquired from the same traffic monitoring site are spatially precisely aligned to ensure successful image fusion. Feature image decomposition transforms the registered image into a more suitable domain for fusion and decomposes its core feature components. Fusion reconstruction involves developing intelligent merging strategies for the different feature components of the two images to be merged.

[0017] The fused image after separating the foreground target is then input into the event detection module. The event detection module automatically detects and outputs the category of candidate traffic events and related event information based on the input fused image. Finally, a multi-level verification module for preventing false positives is constructed, which performs step-by-step verification and filtering through multiple evidence chains, including low-level feature verification, physical law verification, contextual consistency verification, and historical data verification. This allows for in-depth verification of the initially detected candidate traffic events, outputting the final traffic event category and the confidence level for preventing false positives.

[0018] The event detection module is an existing system function. It can accept continuous images or video streams as input, as well as images that have been fused and enhanced, or images that have not been enhanced. The event detection module will perform event detection on the input images in real time and output the category of the traffic event, the confidence level (confidence_pre), and related event information. The related event information can include conventional traffic event information such as object information, trajectory information, and scene information.

[0019] In achieving accurate foreground target separation using an adaptive background modeling method, the foreground target region in the initial fused image is preserved, while the remaining regions are masked. Specifically, the initial fused image is used to separate moving foregrounds through background modeling, which can highlight moving objects or areas with abnormal temperatures, while the remaining regions are masked.

[0020] The aforementioned video traffic incident detector's anti-false alarm system generates enhanced multimodal data through advanced image fusion technology to effectively cope with complex environmental interference such as strong light, backlight, and rain and snow. This provides high-quality data to reduce false alarms and improve the accuracy of judging complex traffic incidents. Then, by using an adaptive background modeling method on the multimodal data, it achieves accurate foreground target separation, highlights moving areas, and suppresses background noise, thereby reducing false alarms. Finally, it combines multi-level progressively refined verification to further reduce false alarms and missed alarms, achieving the goal of significantly reducing the false alarm and missed alarm rates of traffic incidents.

[0021] In one embodiment, further, image preprocessing may specifically include image denoising, contrast enhancement, radiometric correction, and nonlinear correction.

[0022] It is understandable that the image denoising stage aims to prepare high-quality, standardized input data for fusion. Infrared images typically contain a lot of noise (such as shot noise and fixed-pattern noise), so existing nonlocal mean filtering algorithms can be used for image denoising, smoothing out noise while preserving important thermal target edges. For visible light images, the main focus is on dealing with Gaussian noise and salt-and-pepper noise, which can be addressed using existing bilateral filtering methods.

[0023] In the contrast enhancement stage, infrared images have a narrow dynamic range and low contrast. Therefore, existing adaptive histogram equalization (AHE) can be used to enhance their contrast, highlighting targets with small temperature differences. For visible light images, adjustments can be made according to lighting conditions. Local contrast enhancement can be performed on visible light images in backlit scenes to restore details in dark areas. Radiometric correction and nonlinear correction are mainly used for infrared images to eliminate fixed-pattern noise caused by factors such as lens thermal radiation and detector non-uniformity, thereby correcting the linear relationship between the grayscale values ​​of the infrared image and the true temperature.

[0024] In one embodiment, further, image registration spatially aligns the visible light image and the infrared image through scale-invariant feature transformation.

[0025] Image registration is essentially spatial alignment of images. Since infrared and visible light cameras differ in position, angle, and focal length, pixel-level alignment is necessary. Feature-based methods can be used, such as using SIFT (Scale Invariant Feature Transform) to extract common feature points from both images, then calculating the homography matrix between these common feature points to complete the perspective transformation.

[0026] In one embodiment, the feature image decomposition further includes: The registered visible light and infrared images are converted from the spatial domain to the frequency domain. In the frequency domain, the registered visible light and infrared images are decomposed into low-frequency components, high-frequency components, and salient feature components with different characteristics.

[0027] It is understandable that in the feature image decomposition stage, the first step is to select the transform domain, which transforms the registered visible light and infrared images from the spatial domain to the frequency domain, enabling feature separation at different resolutions. Then, feature decomposition is performed: in the transform domain, the registered visible light and infrared images are decomposed into low-frequency components, high-frequency components, and salient feature components with different characteristics. The low-frequency components represent approximate information and low-frequency background of the image. The high-frequency components represent high-frequency information such as edges and textures. The salient feature components are the separately extracted thermal target regions (i.e., highlighted areas) in the infrared image and texture details in the visible light image.

[0028] In one embodiment, further, the fusion reconstruction is performed by fusing the low-frequency components, high-frequency components, and salient targets in the visible light image and the infrared image with salient targets through low-frequency component fusion rules, high-frequency component fusion rules, and salient target fusion rules. The fused image is then transformed back into the spatial domain to obtain a preliminary fused image.

[0029] It is understandable that the fusion and reconstruction stage includes low-frequency component fusion rules, high-frequency component fusion rules, and salient target fusion rules. Low-frequency components primarily contain background energy information; therefore, the low-frequency component fusion rule uses a weighted average selection rule, assigning higher weights to regions with larger variances, which typically contain more information, ensuring a good visual background in the fused image while suppressing noise. High-frequency components contain edge and texture details; therefore, the high-frequency component fusion rule uses a local energy-based rule, selecting pixel values ​​with higher local energy to retain the sharpest edges, maximizing the preservation of rich textures from the visible light image and prominent target contours from the infrared image. For salient thermal target regions extracted from the infrared image, the salient target fusion rule directly replaces the pixel values ​​or gradient information of the thermal target region in the fused image, ensuring that the thermal target is clearly visible in the fused image and unaffected by background interference. Thus, the low-frequency and high-frequency components, fused according to the above three fusion rules, are recombined through corresponding inverse transformations and transformed back to the spatial domain to obtain the preliminary fused image.

[0030] In one embodiment, low-level feature verification includes physical size reasonableness checks, multimodal sensor consistency checks, basic motion continuity checks, and image quality appearance reasonableness checks.

[0031] It is understandable that in the false positive prevention module, low-level feature verification is based on the most intuitive and easily computed low-level features of the target, removing the most unreasonable false positives. The verification content and rejection logic of low-level feature verification can be specifically as follows: Physical size reasonableness check is to check whether the pixel size of the target (such as a car, truck, or pedestrian) is within a reasonable physical range based on the target category detected in the candidate traffic event.

[0032] Multimodal sensor consistency check is performed during image fusion to check whether the target types identified by the infrared camera and the visible light camera are the same or compatible. For example, if the visible light camera identifies a vehicle, the infrared camera should also detect a heat source to determine that the target is the same or compatible.

[0033] The basic motion continuity check combines the detection results of the previous few frames of the current frame to check whether the positional change of the target in the current frame is smooth.

[0034] Image quality appearance plausibility check analyzes the sharpness, contrast, and edge strength of target area images in candidate traffic events. A realistic foreground target should have a relatively clear outline. Images of target areas that are blurry, have a low signal-to-noise ratio, or whose texture features are mismatched (e.g., texture features that resemble shadows, water stains, or changes in light and shadow) will be rejected from the check.

[0035] In the false positive prevention module, the physical law verification is based on fundamental laws of physics to verify the rationality of the target's trajectory, including kinematic rationality verification and trajectory physical consistency verification. Kinematic rationality verification analyzes the short-term trajectory of the target in a candidate traffic event, calculating its instantaneous velocity and acceleration. If the maximum velocity or acceleration violates the specifications, the target's motion is judged to be discontinuous, and the kinematic rationality verification fails. Trajectory physical consistency verification analyzes whether the target's trajectory in the candidate traffic event conforms to the geometric constraints (such as curvature) and dynamic laws (such as centrifugal force) of the road. If not, the trajectory physical consistency verification fails.

[0036] In the anti-false positive module, context consistency verification, also known as scene rationality verification, includes road structure consistency verification, environmental context matching verification, and multi-target interaction rationality verification. Road structure consistency verification compares the position and orientation of targets in candidate traffic events with high-precision maps or real-time perceived static elements (such as lane lines, shoulders, and traffic signs). If the target's position and orientation are unreasonable, the road structure consistency verification fails.

[0037] Contextual matching verification combines environmental information such as weather and lighting to verify the rationality of an event. For example, if a vehicle skids frequently in dry and sunny weather, but the trajectory of the target does not conform to the existing physical model of skid, the system will reduce the confidence of the event or refuse to report the event.

[0038] Multi-objective interaction rationality verification involves analyzing the interaction between a target and other traffic participants (such as vehicles and pedestrians) in a candidate traffic event to verify the event's rationality. For example, if a report shows a "collision" between two vehicles, but the trajectories of the two vehicles show that neither changed speed or direction after the "collision," which clearly does not conform to collision dynamics, the event should not be reported.

[0039] In the false alarm prevention module, historical data verification uses historical big data for intelligent pattern matching to identify abnormal and false alarm patterns, including false alarm pattern library matching, behavioral pattern consistency verification, and statistical anomaly detection.

[0040] False alarm pattern matching: The system maintains a continuously updated false alarm pattern library. When a newly detected candidate traffic event occurs, its features (such as location, size, movement, and appearance) are matched against known false alarm patterns in the library. If the current candidate traffic event is highly similar to a known false alarm pattern that has frequently caused false alarms in the past (for example, at a specific time each day, the shadow of a bridge pier might be mistaken for a stationary vehicle), the system tends to reject the current candidate traffic event.

[0041] Behavioral pattern consistency verification: The behavior (such as speed and trajectory) of the target in the candidate traffic event is compared with the typical historical behavioral patterns of the target's location and time period. If a candidate traffic event reporting "congestion" suddenly appears on a normally smooth road segment, but the candidate traffic event only lasts for 2 to 3 frames, and there has never been a real congestion with a similar pattern in the historical data, the system considers it an accidental false positive and refuses to report the candidate traffic event.

[0042] Statistical anomaly detection: Based on historical statistical data, a normal distribution model is established to determine the frequency of events at different locations and times. If the probability of a current candidate traffic event is significantly lower than the historical statistical anomaly threshold, it is considered a statistical anomaly, and the system refuses to report the current candidate traffic event.

[0043] The system's final output is the category of the traffic event and the confidence level for preventing false positives. The confidence level for preventing false positives is the product of the event confidence level (confidence_step) verified step by step and the confidence level (confidence_pre) output by the event detection module. The implementation process of the false positive module is as follows: Figure 3 As shown.

[0044] Each module in the aforementioned video traffic incident detector's false alarm prevention system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of a device with data processing capabilities, or stored in software within the memory of the aforementioned device, so that the processor can call and execute the corresponding operations of each module. The aforementioned device can be, but is not limited to, various types of computer equipment already existing in the art.

[0045] In one embodiment, such as Figure 4 As shown, a method for preventing false alarms in a video traffic incident detector is provided, which may include the following steps S10 to S16: S10, synchronously acquire visible light video stream and infrared thermal imaging video stream collected by traffic monitoring equipment; S12: After performing image enhancement processing on the visible light video stream and the infrared thermal imaging video stream, adaptive background modeling is used to separate the foreground target and generate fused and enhanced image data; the image enhancement processing includes image preprocessing, image registration, feature image decomposition and fusion reconstruction. S14, automatically detect and output the categories of candidate traffic events and related event information based on the input fused image; S16 performs multi-evidence chain verification and screening of candidate traffic events from coarse to fine, and outputs the final traffic event category and the confidence level to prevent false judgment. Multi-evidence chain verification and screening includes low-level feature verification, physical law verification, context consistency verification and historical data verification.

[0046] The aforementioned method for preventing false alarms in video traffic incident detectors generates enhanced multimodal data through advanced image fusion technology to effectively cope with interference from complex environments such as strong light, backlight, and rain and snow. This provides high-quality data to reduce false alarms and improve the accuracy of judgments on complex traffic incidents. Then, by using an adaptive background modeling method on the multimodal data, accurate foreground target separation is achieved, highlighting the moving area and suppressing background noise, thereby reducing false alarms. Finally, multi-level progressively refined verification is combined to further reduce false alarms and missed alarms, achieving the goal of significantly reducing the false alarm and missed alarm rates of traffic incidents.

[0047] In one embodiment, image preprocessing includes image denoising, contrast enhancement, radiometric correction, and nonlinear correction.

[0048] In one embodiment, image registration spatially aligns visible light and infrared images using scale-invariant feature transformation.

[0049] In one embodiment, low-level feature verification includes physical size reasonableness checks, multimodal sensor consistency checks, basic motion continuity checks, and image quality appearance reasonableness checks. Verification of physical laws includes verification of kinematic rationality and verification of trajectory physical consistency; Context consistency verification includes road structure consistency verification, environmental context matching verification, and multi-objective interaction rationality verification; Historical data verification includes false alarm pattern matching, behavioral pattern consistency verification, and statistical anomaly detection.

[0050] It is understood that the specific limitations of the above-mentioned video traffic incident detector's false judgment prevention method can be found in the corresponding limitations of the video traffic incident detector's false judgment prevention system mentioned above, and will not be repeated here.

[0051] It should be understood that, although Figure 4 The steps are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are executed; they can be performed in other orders. Figure 4 At least some of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0052] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus DRAM (RDRAM), and interface DRAM (DRDRAM), etc.

[0053] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0054] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of protection of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and all such modifications and improvements fall within the scope of protection of the present invention.

Claims

1. A system for preventing false alarms in a video traffic incident detector, characterized in that, include: The image acquisition module is used to simultaneously acquire visible light video streams and infrared thermal imaging video streams collected by traffic monitoring equipment; The image enhancement module is used to perform image enhancement processing on visible light video streams and infrared thermal imaging video streams, and then use adaptive background modeling to separate foreground targets and generate fused and enhanced image data. Image enhancement processing includes image preprocessing, image registration, feature image decomposition, and fusion reconstruction; The event detection module is used to automatically detect and output the categories of candidate traffic events and related event information based on the input fused image; The false positive prevention module is used to perform multi-evidence chain verification and screening of candidate traffic events from coarse to fine, and output the final traffic event category and false positive confidence. The multi-evidence chain verification and screening includes low-level feature verification, physical law verification, context consistency verification and historical data verification.

2. The anti-false alarm system for video traffic incident detectors according to claim 1, characterized in that, Image preprocessing includes image denoising, contrast enhancement, radiometric correction, and nonlinear correction.

3. The anti-false alarm system for video traffic incident detectors according to claim 1 or 2, characterized in that, Image registration spatially aligns visible light and infrared images using scale-invariant feature transformation.

4. The anti-false alarm system for video traffic incident detectors according to claim 1 or 2, characterized in that, Feature image decomposition includes: The registered visible light and infrared images are converted from the spatial domain to the frequency domain. In the frequency domain, the registered visible light and infrared images are decomposed into low-frequency components, high-frequency components, and salient feature components with different characteristics.

5. The anti-false alarm system for video traffic incident detectors according to claim 4, characterized in that, Fusion reconstruction involves fusing low-frequency components, high-frequency components, and salient targets from visible light and infrared images using low-frequency component fusion rules, high-frequency component fusion rules, and salient target fusion rules. The fused image is then transformed back into the spatial domain to obtain a preliminary fused image.

6. The anti-false alarm system for video traffic incident detectors according to claim 1, characterized in that, Low-level feature verification includes physical size rationality checks, multimodal sensor consistency checks, basic motion continuity checks, and image quality appearance rationality checks. Verification of physical laws includes verification of kinematic rationality and verification of trajectory physical consistency; Context consistency verification includes road structure consistency verification, environmental context matching verification, and multi-objective interaction rationality verification; Historical data verification includes false alarm pattern matching, behavioral pattern consistency verification, and statistical anomaly detection.

7. A method for preventing false judgments in a video traffic incident detector, characterized in that, Including the following steps: Simultaneously acquire visible light video streams and infrared thermal imaging video streams collected by traffic monitoring equipment; After image enhancement processing of visible light video stream and infrared thermal imaging video stream, adaptive background modeling is used to separate foreground targets and generate fused and enhanced image data. Image enhancement processing includes image preprocessing, image registration, feature image decomposition, and fusion reconstruction; The system automatically detects and outputs the categories of candidate traffic events and related event information based on the input fused image. Candidate traffic events are subjected to progressive verification and screening using a multi-evidence chain from coarse to fine, and the final category of the traffic event and the confidence level for preventing false positives are output. The progressive verification and screening of the multi-evidence chain includes low-level feature verification, physical law verification, contextual consistency verification, and historical data verification.

8. The method for preventing false judgments in a video traffic incident detector according to claim 7, characterized in that, Image preprocessing includes image denoising, contrast enhancement, radiometric correction, and nonlinear correction.

9. The method for preventing false judgments in a video traffic incident detector according to claim 7 or 8, characterized in that, Image registration spatially aligns visible light and infrared images using scale-invariant feature transformation.

10. The method for preventing false judgments in a video traffic incident detector according to claim 7, characterized in that, Low-level feature verification includes physical size rationality checks, multimodal sensor consistency checks, basic motion continuity checks, and image quality appearance rationality checks. Verification of physical laws includes verification of kinematic rationality and verification of trajectory physical consistency; Context consistency verification includes road structure consistency verification, environmental context matching verification, and multi-objective interaction rationality verification; Historical data verification includes false alarm pattern matching, behavioral pattern consistency verification, and statistical anomaly detection.

Images