A raspberry-pi-based multi-sensor fusion intelligent border monitoring system and method

Abstract

The application discloses a kind of based on raspberry pi's multi-sensor fusion intelligent boundary monitoring system and method, belong to boundary safety monitoring technical field.The system is deployed in raspberry pi platform, including main control module, multi-sensor detection module (infrared, camera, radiation sensor), multi-modal detection algorithm module (fusion background reduction, three frame difference method and deep learning target identification), adaptive environment regulation module, event correlation analysis module and alarm response module.System can be stably operated on embedded device through multi-sensor information fusion, intelligent algorithm and resource optimization design, significantly improves detection accuracy, environmental adaptability and reduces false alarm rate, realizes the intelligent identification and radiation monitoring function of personnel and vehicle.The application is especially suitable for nuclear facilities and other special safety places that need radiation monitoring.

Description

Technical Field

[0001] This invention relates to the technical field of border security monitoring, specifically to a multi-sensor fusion intelligent border monitoring system and method based on Raspberry Pi. Background Technology

[0002] Border security monitoring is a key technology for ensuring the security of important areas. Traditional border monitoring systems mainly rely on a single sensor (such as an infrared sensor or a camera), which has the following problems: 1. Low detection accuracy, prone to false alarms and missed alarms; 2. It cannot adapt to complex environmental changes, especially in scenes with large changes in lighting, where the detection performance is poor; 3. It lacks intelligent target recognition capabilities and cannot distinguish between different target types such as people and vehicles; 4. It lacks radiation monitoring capabilities, failing to meet the safety monitoring needs of special locations such as nuclear facilities; 5. The system consumes a lot of resources and is difficult to run stably on embedded devices.

[0003] While some existing technologies employ multi-sensor fusion monitoring systems, most suffer from problems such as complex algorithms, poor real-time performance, high false alarm rates, and inability to adapt to environmental changes. Therefore, there is an urgent need for a system capable of accurate, real-time, and stable boundary monitoring. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of low detection accuracy, high false alarm rate, inability to adapt to environmental changes, and lack of intelligent target recognition in the existing technology. It proposes a multi-sensor fusion intelligent boundary monitoring system and method based on Raspberry Pi to achieve accurate, real-time, and adaptive intelligent boundary monitoring on resource-constrained embedded platforms.

[0005] To achieve the above objectives, this invention proposes a multi-sensor fusion intelligent boundary monitoring system based on Raspberry Pi. The system is deployed on a Raspberry Pi hardware platform and includes: The main control module is used to manage the various detection threads and system resources; The multi-sensor detection module includes an infrared motion sensor, a camera, and a radiation sensor, which are used to collect motion trigger signals, video image data, and radiation data in the boundary area, respectively. The multimodal detection algorithm module includes a background subtraction algorithm, a three-frame difference motion detection algorithm, and a deep learning-based target recognition algorithm, which are used to perform multi-level motion detection and target recognition on the collected data. An adaptive environment adjustment module is used to automatically adjust the camera's exposure parameters according to the ambient brightness. The event correlation analysis module is used to perform intelligent correlation analysis on detected events to reduce false alarms; The alarm response module is used to trigger an alarm and automatically take pictures to record when an abnormal situation is detected.

[0006] Furthermore, the multimodal detection algorithm module includes: The background subtraction unit, using MOG2 or KNN background subtractors, is used to detect moving objects in static scenes. The three-frame difference unit is used to detect fast-moving objects by calculating the differences between three consecutive frames of images. The fusion detection unit performs a logical OR operation on the background subtraction result and the three-frame difference result; The target recognition unit, when detecting a moving area, invokes the target recognition model to identify people and vehicles. The recognition categories include people, cars, motorcycles, buses, and trucks.

[0007] Furthermore, the adaptive environment adjustment module includes: A brightness detection unit is used to calculate the average brightness value of the image; The exposure adjustment unit automatically adjusts the camera's exposure mode, exposure compensation value, and brightness parameters based on the average brightness value. The system includes a low brightness threshold, a high brightness threshold, and an optimal brightness range. The exposure adjustment unit is configured to: increase exposure and brightness when the average brightness is lower than the low brightness threshold; decrease exposure and brightness when the average brightness is higher than the high brightness threshold; and maintain automatic exposure mode when the average brightness is within the optimal brightness range.

[0008] Furthermore, the event correlation analysis module includes: The spatial association unit determines whether the events are the same based on the distance between the center points of motion. If the distance is less than a preset distance threshold, the events are determined to be the same. The time-related unit sets a cooldown period for events to avoid recording the same event repeatedly in a short period of time; The priority determination unit determines the event type according to the rule that personnel detection takes precedence over vehicle detection.

[0009] Furthermore, the radiation sensor is connected via serial port through the Modbus RTU communication protocol, enabling real-time monitoring of radiation dose rate, count rate, and smoothed dose rate, and triggering an alarm when the radiation level exceeds a preset threshold.

[0010] Furthermore, the system employs a multi-threaded concurrent design, including: The motion detection thread is used to perform video motion detection and object recognition; The infrared detection thread is used to perform infrared sensor detection. The radiation detection thread is used to perform radiation data acquisition and anomaly detection. Each thread runs independently, and the security of data access is ensured through a thread lock mechanism.

[0011] Furthermore, the alarm response module includes: An audio alarm unit is used to play warning voice files, and the number of alarm playbacks is configurable. The photo recording unit is used to automatically take multiple high-resolution photos in response to the detection of people or vehicles, and add timestamps and detection information to the photos; The license plate recognition unit detects the license plate area of ​​the detected vehicle and identifies blue-background license plates based on the HSV color space and shape features.

[0012] Furthermore, the system also includes: The resource optimization module is used to reduce system load by at least one of the following methods: reducing motion detection resolution, limiting target recognition frequency, and optimizing preview display. The log management module employs log filtering and cooling mechanisms to reduce duplicate log output.

[0013] This invention also proposes a multi-sensor fusion intelligent boundary monitoring method based on Raspberry Pi, comprising the following steps: S1: Initialize the system, start the camera, infrared sensor and radiation sensor, and load the target detection model; S2: Motion detection is performed using a multimodal detection algorithm, which includes background subtraction and fusion detection using the three-frame difference method; S3: When a moving area is detected, the target recognition model is invoked to perform target recognition, which includes people and vehicles; S4: Automatically adjusts camera exposure parameters based on ambient brightness; S5: Perform intelligent correlation analysis on detected events to determine whether they are the same event; S6: When a person or vehicle is detected, an alarm is triggered and a photo is automatically taken to record the event. S7: Monitors radiation levels in real time and triggers an alarm when the level exceeds a preset threshold.

[0014] Furthermore, the multimodal detection in step S2 includes: A background subtractor is used to detect moving objects in a static scene; The three-frame difference method is used to detect fast-moving objects; The background subtraction result and the three-frame difference result are fused together.

[0015] The beneficial effects of this invention are: 1. Optimized implementation based on Raspberry Pi platform: The system is designed specifically for the Raspberry Pi embedded hardware platform. Through resource optimization modules and multi-threaded concurrency design, it solves the problem of stable operation of multimodal intelligent algorithms under limited computing resources.

[0016] 2. Multimodal detection algorithm fusion: By combining background subtraction, three-frame difference method and deep learning-based target recognition, the detection accuracy and target recognition ability are significantly improved, and multiple target types such as people, cars, and motorcycles can be accurately distinguished.

[0017] 3. Adaptive environmental adjustment: The camera parameters are automatically adjusted according to the ambient brightness to ensure that the system can work stably under different lighting conditions during the day and night, thus improving the system's environmental adaptability.

[0018] 4. Intelligent event correlation analysis: Through spatial correlation and temporal correlation (cooling-off period) mechanisms, false alarms and duplicate records are effectively reduced, improving the reliability of the system.

[0019] 5. Multi-sensor fusion: Combining infrared sensors, cameras, and radiation sensors, it enables multi-dimensional acquisition and analysis of motion trigger signals, video images, and radiation data, making it particularly suitable for special locations such as nuclear facilities.

[0020] 6. Multi-threaded concurrent design: Each detection module runs independently without blocking each other, improving the system's real-time performance and response speed.

[0021] 7. Resource optimization: By implementing measures such as low-resolution detection, detection frequency control, and preview optimization, the system resource consumption is reduced, enabling stable operation on embedded devices (such as Raspberry Pi).

[0022] 8. Radiation monitoring function: It integrates a radiation sensor, which can monitor the radiation level in real time and automatically alarm when the threshold is exceeded, meeting the safety monitoring needs of special places such as nuclear facilities.

[0023] The features and advantages of the present invention will be described in detail through embodiments and in conjunction with the accompanying drawings. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall architecture of the intelligent boundary monitoring system of the present invention; Figure 2 This is a flowchart of the multimodal detection algorithm module of the present invention; Figure 3 This is a flowchart of the adaptive environment adjustment module of the present invention; Figure 4 This is a flowchart of the event correlation analysis module of the present invention; Figure 5 This is a schematic diagram of the multi-threaded concurrent design of the system of this invention; Figure 6 This is a flowchart of the intelligent boundary monitoring method of the present invention. Detailed Implementation

[0025] See Figures 1-6 This invention discloses a multi-sensor fusion intelligent boundary monitoring system and method based on Raspberry Pi. The core architecture of the system mainly includes: a main control module, a multi-sensor detection module, a multi-modal detection algorithm module, an adaptive environment adjustment module, an event correlation analysis module, and an alarm response module. All modules work collaboratively on the Raspberry Pi hardware platform. The main control module acts as the system's central hub, responsible for scheduling and managing all other functional modules and system resources.

[0026] The main structure and connections of this invention are as follows: The main control module serves as the system core, interacting with each functional module via data and commands. In the multi-sensor detection module, the infrared motion sensor, camera, and radiation sensor are connected according to the Raspberry Pi hardware specifications. For example, the infrared sensor is connected to a designated GPIO pin (such as GPIO17), the camera uses the official Raspberry Pi camera module, and the radiation sensor is connected via a serial port (such as / dev / ttyUSB0). The multimodal detection algorithm module receives video data from the camera; the adaptive environment adjustment module sends control parameters to the camera; and the event correlation analysis module receives target information and outputs the judgment result to the alarm response module.

[0027] Working process of this invention: The present invention relates to a multi-sensor fusion intelligent boundary monitoring system and method based on Raspberry Pi, which is described in conjunction with the accompanying drawings during operation.

[0028] Example 1 like Figure 1 As shown, the present invention includes a main control module, a multi-sensor detection module, a multi-modal detection algorithm module, an adaptive environment adjustment module, an event correlation analysis module, and an alarm response module. The system is deployed on a Raspberry Pi hardware platform.

[0029] The main control module is responsible for the overall coordination and resource management of the system. Upon system startup, the main control module first initializes the camera (using the Picamera2 library, configured with a high-resolution mainstream 1920x1080 for image capture and a low-resolution 640x480 stream for real-time detection), the infrared sensor (connected to the GPIO17 pin), and the radiation sensor (connected via serial port / dev / ttyUSB0, using the Modbus RTU protocol). Then, it loads the target detection model (e.g., YOLOv8n) and creates three independent detection threads: a motion detection thread, an infrared detection thread, and a radiation detection thread.

[0030] The multi-sensor detection module includes: (1) Infrared motion sensor: A PIR (passive infrared) sensor is used, connected to the GPIO17 pin of the Raspberry Pi, to detect human motion and collect motion trigger signals. A software filtering mechanism is used, requiring three consecutive motion detections before triggering, with a minimum trigger interval of 2 seconds to reduce false alarms.

[0031] (2) Camera: The Raspberry Pi official camera module is used, which supports the libcamera interface. Dual-stream mode is configured: the mainstream 1920x1080 is used for high-resolution photo recording, and the low-resolution 640x480 is used for real-time motion detection to reduce computational complexity and for acquiring video image data.

[0032] (3) Radiation Sensor: A FUS-06 radiation sensor is used, connected via serial port, using the Modbus RTU communication protocol. The sensor address is 0x01, and the baud rate is 9600. It can read the radiation dose rate (register 0x0001, unit μSv / h), count rate (register 0x0005, unit CPS), and smoothed dose rate (register 0x0013, unit μSv / h) for radiation data acquisition. The sensor requires a 30-second warm-up time to ensure stable readings.

[0033] like Figure 2 As shown, the workflow of the multimodal detection algorithm module is as follows: (1) Background subtraction: The MOG2 background subtractor (200 historical frames, variance threshold 25, no shadow detection) is used to process the grayscale image to obtain the background mask.

[0034] (2) Three-frame difference method: Maintain three consecutive frames of images (current frame, previous frame, and the frame before that), calculate the first frame difference (the absolute difference between the current frame and the previous frame) and the second frame difference (the absolute difference between the previous frame and the frame before that), perform a logical AND operation on the two frame differences to obtain the three-frame difference result.

[0035] (3) Fusion detection: The background subtraction result and the three-frame difference result are logically ORed to combine the advantages of the two methods. Background subtraction is suitable for motion detection in static scenes, while the three-frame difference method is suitable for detecting fast-moving objects.

[0036] (4) Morphological processing: Opening (removing small noise) and closing (filling holes) operations are performed on the fused mask, using a 5x5 elliptic kernel.

[0037] (5) Contour analysis: Extract motion contours, filter contours with areas that are too small (less than 150 pixels) or too large (greater than 40,000 pixels), and calculate the motion center point.

[0038] (6) Target Recognition: When a valid moving area is detected and the camera is not moving, the target recognition model (e.g., YOLO model) is invoked for target recognition. The detection interval of this model is set to 1.5 seconds, the input resolution is 256x256 (Raspberry Pi optimized), and the confidence threshold is 0.5. The recognition categories include: people (class_id=0), cars (class_id=2), motorcycles (class_id=3), buses (class_id=5), and trucks (class_id=7).

[0039] like Figure 3 As shown, the workflow of the adaptive environment adjustment module is as follows: (1) Brightness detection: Check the ambient brightness every 60 seconds and calculate the average brightness value of the image (calculate the mean value after converting the BGR image to grayscale).

[0040] (2) Brightness judgment: Determine the ambient lighting conditions based on the average brightness value: - Low brightness (<40): Increase exposure compensation value to 2.0, increase brightness to 0.3, and enable automatic exposure mode.

[0041] - High brightness (>180): Reduce exposure compensation value to -2.0, reduce brightness to -0.2, and enable automatic exposure mode.

[0042] - Optimal brightness (80-150): Keep the automatic exposure mode and do not adjust other parameters.

[0043] (3) Parameter application: Apply the adjusted parameters through the set_controls method of Picamera2.

[0044] like Figure 4 As shown, the workflow of the event correlation analysis module is as follows: (1) Spatial association: When a new motion event is detected, the Euclidean distance between the current motion center point and the center point of the historical event is calculated. If the distance is less than a preset distance threshold (e.g., 50 pixels), it is determined to be the same event, and the event status is updated; otherwise, a new event is created.

[0045] (2) Time correlation: A cooldown period (e.g., 4 seconds) is set for each event. During the cooldown period, the photo will not be triggered repeatedly to avoid recording the same event repeatedly in a short period of time. The event will be automatically cleared after the last update time exceeds the cooldown period.

[0046] (3) Priority determination: The event type and trigger type are determined according to the priority rule of personnel detection > vehicle detection > general movement.

[0047] (4) Photo capture strategy: A 1-second cooldown time is used for personnel detection, and a standard cooldown time (5 seconds) is used for other types. A maximum of 2 photos are taken for each event, with a 1.5-second interval between photos.

[0048] like Figure 5 As shown, the system adopts a multi-threaded concurrent design: (1) Motion detection thread: performs video motion detection and target recognition, with a loop cycle of approximately 0.1 seconds.

[0049] (2) Infrared detection thread: Performs infrared sensor detection with a check interval of 0.5 seconds and uses software filtering to reduce false alarms.

[0050] (3) Radiation detection thread: performs radiation data acquisition, updates every 2 seconds, and monitors radiation levels in real time.

[0051] Each thread protects shared data through a thread lock (threading.Lock) to ensure thread safety.

[0052] The alarm response module includes: (1) Audio alarm: When an anomaly is detected, a warning audio file (warning.mp3) is played. The number of times it is played can be configured (e.g., 3 times). It is implemented using the pygame library.

[0053] (2) Photo recording: In response to the detection of people or vehicles, automatically take 3 high-resolution photos (1920x1080) at 0.3-second intervals. Add timestamps and detection information (detection type, photo number) to the photos, save them in JPEG format with 95% quality.

[0054] (3) License plate recognition: For the detected vehicle, extract the vehicle area, convert it to HSV color space, detect the blue area (HSV range: 100-130, 50-255, 50-255), and identify the license plate area (width-to-height ratio 2.0-4.0) through morphological operations and contour analysis.

[0055] Resource optimization measures: (1) Use low-resolution frames (640x480) for motion detection to reduce computational complexity.

[0056] (2) Limit the target identification frequency to a 1.5-second interval to balance real-time performance and system load.

[0057] (3) Optimize preview display: The preview resolution is fixed at 640x480, and the number of detection boxes drawn is limited (maximum 2 people and 2 vehicles).

[0058] (4) Log management: Use log filters to filter duplicate DEBUG logs and set a log cooldown time (2 seconds) to avoid frequent recording.

[0059] Example 2 This embodiment provides a multi-sensor fusion intelligent boundary monitoring system and method based on Raspberry Pi, including the following steps: S1: System Initialization Start the camera, configure dual-stream mode (mainstream 1920x1080, low-resolution stream 640x480), set the frame rate to 30fps, and enable automatic exposure. Initialize the infrared sensor (GPIO17), and set the detection threshold and interval. Initialize the radiation sensor (serial port / dev / ttyUSB0), and wait 30 seconds for warm-up. Load the target detection model (e.g., YOLOv8n), set the confidence threshold to 0.5, and the input resolution to 256x256. Initialize the background subtractor (MOG2) and create three detection threads.

[0060] S2: Multimodal motion detection Acquire low-resolution frames (640x480), convert them to grayscale, and apply medium-level filtering for noise reduction. Perform background subtraction to obtain a background mask. Perform three-frame difference calculation, finding the intersection of three consecutive frames. Fuse the two detection results (using logical OR operation) and perform morphological processing. Extract motion contours, filter invalid contours, and calculate the motion center point. Determine if it's camera movement (motion area > 80%).

[0061] S3: Target Recognition When valid motion is detected that is not camera movement, determine if the target recognition condition is met (following the last detection >= 1.5 seconds). Invoke the target recognition model (e.g., the loaded YOLOv8n model) to perform whole-image detection, identifying people and vehicles. Filter low-confidence results (<0.5), and record the location and category of the detected targets.

[0062] S4: Adaptive Environmental Adjustment The ambient brightness is checked every 60 seconds, and the average image brightness is calculated. Camera parameters are adjusted based on the brightness value: increase exposure in low light, decrease exposure in high light, and maintain optimal brightness in automatic mode.

[0063] S5: Event Correlation Analysis Calculate the distance between the current motion center point and the center points of historical events to determine if they are the same event (distance < 50 pixels). If they are the same event, update the event status; otherwise, create a new event. Determine the event type (person / vehicle / motion) based on the detection results and process it according to priority rules.

[0064] S6: Alarm Response When a person or vehicle is detected, determine if the conditions for taking a photo are met (cooldown time, photo limit). Take 3 high-resolution photos (0.3-second intervals), add timestamps and detection information. Play a warning voice message (playback count is configurable, e.g., 3 times). If it is a vehicle, attempt to identify the license plate area.

[0065] S7: Radiation Monitoring Radiation data (dose rate, count rate, smoothed dose rate) is read every 2 seconds. The radiation level is checked to see if it exceeds the thresholds (warning threshold 0.5 μSv / h, alarm threshold 2.0 μSv / h). If the thresholds are exceeded, an alarm is triggered and a photograph is taken for recording.

[0066] The system runs continuously, with each thread working independently, and data consistency is ensured through thread locks. When a stop signal is received, all sensors and threads are safely shut down, releasing system resources.

[0067] The above embodiments are illustrative of the present invention and are not intended to limit the present invention. Any simple modifications to the present invention are within the scope of protection of the present invention.

Claims

1. A multi-sensor fusion intelligent boundary monitoring system based on Raspberry Pi, characterized in that, The system is deployed on a Raspberry Pi hardware platform and includes: The main control module is used to manage the various detection threads and system resources; The multi-sensor detection module includes an infrared motion sensor, a camera, and a radiation sensor, which are used to collect motion trigger signals, video image data, and radiation data in the boundary area, respectively. The multimodal detection algorithm module includes a background subtraction algorithm, a three-frame difference motion detection algorithm, and a deep learning-based target recognition algorithm, which are used to perform multi-level motion detection and target recognition on the collected data. An adaptive environment adjustment module is used to automatically adjust the camera's exposure parameters according to the ambient brightness. The event correlation analysis module is used to perform intelligent correlation analysis on detected events to reduce false alarms; The alarm response module is used to trigger an alarm and automatically take pictures to record when an abnormal situation is detected.

2. The intelligent boundary monitoring system according to claim 1, characterized in that, The multimodal detection algorithm module includes: The background subtraction unit, using MOG2 or KNN background subtractors, is used to detect moving objects in static scenes. The three-frame difference unit is used to detect fast-moving objects by calculating the differences between three consecutive frames of images. The fusion detection unit performs a logical OR operation on the background subtraction result and the three-frame difference result; The target recognition unit, when detecting a moving area, invokes the target recognition model to identify people and vehicles. The recognition categories include people, cars, motorcycles, buses, and trucks.

3. The intelligent boundary monitoring system according to claim 1, characterized in that, The adaptive environment adjustment module includes: A brightness detection unit is used to calculate the average brightness value of the image; The exposure adjustment unit automatically adjusts the camera's exposure mode, exposure compensation value, and brightness parameters based on the average brightness value. The system includes a low brightness threshold, a high brightness threshold, and an optimal brightness range. The exposure adjustment unit is configured to: increase exposure and brightness when the average brightness is lower than the low brightness threshold; decrease exposure and brightness when the average brightness is higher than the high brightness threshold; and maintain automatic exposure mode when the average brightness is within the optimal brightness range.

4. The intelligent boundary monitoring system according to claim 1, characterized in that, The event correlation analysis module includes: The spatial association unit determines whether the events are the same based on the distance between the center points of motion. If the distance is less than a preset distance threshold, the events are determined to be the same. The time-related unit sets a cooldown period for events to avoid recording the same event repeatedly in a short period of time; The priority determination unit determines the event type according to the rule that personnel detection takes precedence over vehicle detection.

5. The intelligent boundary monitoring system according to claim 1, characterized in that, The radiation sensor is connected via serial port through the Modbus RTU communication protocol and can monitor radiation dose rate, count rate and smoothed dose rate in real time. When the radiation level exceeds a preset threshold, an alarm is triggered.

6. The intelligent boundary monitoring system according to claim 1, characterized in that, The system adopts a multi-threaded concurrent design, including: The motion detection thread is used to perform video motion detection and object recognition; The infrared detection thread is used to perform infrared sensor detection. The radiation detection thread is used to perform radiation data acquisition and anomaly detection. Each thread runs independently, and the security of data access is ensured through a thread lock mechanism.

7. The intelligent boundary monitoring system according to claim 1, characterized in that, The alarm response module includes: An audio alarm unit is used to play warning voice files, and the number of alarm playbacks is configurable. The photo recording unit is used to automatically take multiple high-resolution photos in response to the detection of people or vehicles, and add timestamps and detection information to the photos; The license plate recognition unit detects the license plate area of ​​the detected vehicle and identifies blue-background license plates based on the HSV color space and shape features.

8. The intelligent boundary monitoring system according to claim 1, characterized in that, The system also includes: The resource optimization module is used to reduce system load by at least one of the following methods: reducing motion detection resolution, limiting target recognition frequency, and optimizing preview display. The log management module employs log filtering and cooling mechanisms to reduce duplicate log output.

9. A multi-sensor fusion intelligent boundary monitoring method based on Raspberry Pi, characterized in that, Includes the following steps: S1: Initialize the system, start the camera, infrared sensor and radiation sensor, and load the target detection model; S2: Motion detection is performed using a multimodal detection algorithm, which includes background subtraction and fusion detection using the three-frame difference method; S3: When a moving area is detected, the target recognition model is invoked to perform target recognition, which includes people and vehicles; S4: Automatically adjusts camera exposure parameters based on ambient brightness; S5: Perform intelligent correlation analysis on detected events to determine whether they are the same event; S6: When a person or vehicle is detected, an alarm is triggered and a photo is automatically taken to record the event. S7: Monitors radiation levels in real time and triggers an alarm when the level exceeds a preset threshold.

10. The intelligent boundary monitoring method according to claim 9, characterized in that, The multimodal detection in step S2 includes: A background subtractor is used to detect moving objects in a static scene; The three-frame difference method is used to detect fast-moving objects; The background subtraction result and the three-frame difference result are fused together.

Images