Humanoid filtering algorithm for humanoid smart cameras in security monitoring

Abstract

The present invention discloses a humanoid filtering algorithm for a humanoid intelligent camera in the field of security monitoring, and relates to the technical field of security monitoring. The humanoid filtering algorithm for the humanoid intelligent camera in the field of security monitoring includes the following steps: preparation for humanoid filtering detection, calibration and processing of the detection frame, detection of group one and detection of group two. By taking turns to send group one and group two for inspection in a poor detection manner, the rate of detecting humanoids can be significantly improved, while ensuring that one of every two inspections must be a fixed frame detection. In this way, the problem of people being unable to be detected when they are stationary in the picture can be solved, the false alarm rate is significantly reduced, and the time consumption of detection is effectively reduced. On the basis of the humanoid algorithm, the humanoid detection results at the top of the video picture and the very small BGM results are filtered out, thereby reducing the false alarm rate of humanoids, and the detection distance is increased to 10 meters, effectively improving the detection efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of security monitoring, and in particular to a humanoid filtering algorithm for a humanoid intelligent camera in the field of security monitoring. Background Art

[0002] Artificial neural network is a hot research topic in the field of artificial intelligence since the 1980s. It abstracts the human brain neural network from the perspective of information processing, establishes a simple model, and forms different networks according to different connection methods. In engineering and academia, it is often referred to as neural network or quasi-neural network. Neural network is a computing model composed of a large number of nodes (or neurons) connected to each other. Each node represents a specific output function, called the activation function. The connection between each two nodes represents a weighted value for the signal passing through the connection, called the weight, which is equivalent to the memory of the artificial neural network. The output of the network varies depending on the connection method, weight value and activation function of the network. The network itself is usually an approximation of a certain algorithm or function in nature, or it may be an expression of a logical strategy. In the past ten years, the research work on artificial neural network has been continuously deepened and has made great progress. It has successfully solved many practical problems that are difficult to solve with modern computers in the fields of pattern recognition, intelligent robots, automatic control, prediction and estimation, biology, medicine, and economy, showing good intelligent characteristics.

[0003] The existing humanoid algorithms are mainly based on artificial neural network algorithms for detection. Artificial neural network, referred to as neural network or quasi-neural network, is a mathematical model or computational model that imitates the structure and function of biological neural networks and is used to estimate or approximate functions. Neural networks have shown excellent performance in various scenarios such as voice, pictures, videos, games, etc., but there is a problem that a large amount of data is needed for training to improve accuracy.

[0004] The principle of the current humanoid detection algorithm is: first, all motion areas of the entire screen are detected through the BGM detection algorithm, and then the data of all motion areas are detected by the neural network algorithm in turn. Due to the limited computing power of the main control, the following problems exist in the actual test:

[0005] It takes a long time to detect a frame of human data, up to 2 seconds per frame, resulting in poor real-time detection; the detection distance is too short, and only people within 5 meters can be detected; the algorithm has a false alarm rate of over 80% for columnar objects such as telephone poles, tree trunks, and warning poles; the algorithm does not support the detection of static people, which will lead to missed detection of people. Summary of the invention

[0006] 1. Technical issues to be solved

[0007] In view of the deficiencies in the prior art, the present invention provides a humanoid filtering algorithm for a humanoid intelligent camera in the field of security monitoring, which solves the problems mentioned in the above background technology.

[0008] (II) Technical solution

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: a humanoid filtering algorithm for a humanoid intelligent camera in the field of security monitoring, the humanoid filtering algorithm comprising the following steps:

[0010] Step 1: Preparation for human filtering detection, obtain the results of the BGM motion area, and filter out the motion frame at the top of the video screen and particularly small motion frames.

[0011] Step 2: Calibrate and process the detection frame. The current detection screen width and height are 640*368. Define 4 detection frames by yourself. The x, y coordinates and width and height are:

[0012] Full frame: x=0, y=20, w=640, h=320;

[0013] Middle box: x=20, y=100, w=600, h=120;

[0014] Lower left frame: x=0, y=168, w=600, h=200;

[0015] Lower right frame: x=40, y=168, w=600, h=200;

[0016] The BGM motion frame and the full-frame frame are set as group one, and the middle frame, lower left frame and lower right frame are set as group two for cross detection, and group one and group two are sent for inspection in turn.

[0017] Step 3: Detect group one. First, detect all the moving frames of the BGM in group one. As long as one BGM frame detects a human figure, exit and alarm. Otherwise, if the BGM frame does not detect a human figure, detect the full-frame frame once. If one is detected, exit and alarm. This can significantly increase the rate of detecting a human figure.

[0018] Step 4: Detect group 2. After group 1 is detected, group 2 is detected. Select one frame from the middle frame, lower left frame and lower right frame of group 2 for detection. If a human figure is detected, make a mark, exit and alarm. The next time group 2 is polled, select the frame for detection. If there is no human figure after two consecutive pollings, switch to other frames for detection. If the selected frame does not detect a human figure, select one of the other two frames when group 2 is polled next time. After the frames in group 2 are detected, group 1 is detected. The detection is cyclical. This ensures that one of the two inspections must be a fixed frame detection. This can solve the problem of not being detected when a person is still in the picture, and significantly reduce the missed detection rate.

[0019] Preferably, in step one, a BGM detection algorithm is used to detect the motion area, and the detected motion area is recorded.

[0020] Preferably, in step 1, a neural network algorithm is used to filter the motion frame at the top of the video image and particularly small motion frames.

[0021] Preferably, in step 2, during the process of sending group 1 and group 2 for inspection in turn, a cyclical manner is used for the inspection.

[0022] 5. The humanoid filtering algorithm for a humanoid intelligent camera in the field of security monitoring according to claim 1 is characterized in that the alarm in step 3 and step 4 adopts one or more of sound alarm, light alarm and vibration alarm.

[0023] Preferably, in step three, one box is selected from the middle box, the lower left box and the lower right box of group two for detection, and the selection is performed randomly.

[0024] Preferably, in step three, the detection of the human figure is marked by using a data processing method.

[0025] Preferably, in step three, a cyclic detection is performed until a human figure is detected and an alarm is sounded, and then the detection is stopped.

[0026] (III) Beneficial effects

[0027] The present invention provides a humanoid filtering algorithm for a humanoid intelligent camera in the field of security monitoring. It has the following beneficial effects:

[0028] (1) The human filtering algorithm of the humanoid intelligent camera in the security monitoring field can significantly improve the detection rate of human figures by using a poor detection method to inspect group one and group two in turn, while ensuring that one of every two inspections is a fixed frame detection. This can solve the problem of not being able to detect a person who is stationary in the picture, significantly reduce the missed detection rate, and effectively reduce the time consumption of detection.

[0029] (2) The human figure filtering algorithm of the human figure intelligent camera in the security monitoring field, if a human figure appears at the top of the video screen, it means that the current distance between the human figure and the camera lens is more than 10 meters, so the final detection distance is set to 10 meters as the acceptance standard. Based on the human figure algorithm, the human figure detection results at the top of the video screen and the very small BGM results are filtered out, thereby reducing the false alarm rate of the human figure and increasing the detection distance to 10 meters, effectively improving the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] See also Figure 1 The present invention provides a technical solution: a humanoid filtering algorithm for a humanoid intelligent camera in the field of security monitoring, the humanoid filtering algorithm comprising the following steps:

[0033] Step 1: Preparation for human filtering detection, obtain the results of the BGM motion area, and filter out the motion frame at the top of the video screen and particularly small motion frames.

[0034] Step 2: Calibrate and process the detection frame. The current detection screen width and height are 640*368. Define 4 detection frames by yourself. The x, y coordinates and width and height are:

[0035] Full frame: x=0, y=20, w=640, h=320;

[0036] Middle box: x=20, y=100, w=600, h=120;

[0037] Lower left frame: x=0, y=168, w=600, h=200;

[0038] Lower right frame: x=40, y=168, w=600, h=200;

[0039] The BGM motion frame and the full-frame frame are set as group one, and the middle frame, lower left frame and lower right frame are set as group two for cross detection, and group one and group two are sent for inspection in turn.

[0040] Step 3: Detect group one. First, detect all the moving frames of the BGM in group one. As long as one BGM frame detects a human figure, exit and alarm. Otherwise, if the BGM frame does not detect a human figure, detect the full-frame frame once. If it is detected, exit and alarm. This can significantly increase the rate of detecting human figures.

[0041] Step 4: Detect group 2. After group 1 is detected, group 2 is detected. Randomly select one frame from the middle frame, lower left frame and lower right frame of group 2 for detection. If a human figure is detected, make a mark, exit and alarm. The next time group 2 is polled, select the frame for detection. If there is no human figure after two consecutive pollings, switch to other frames for detection. If the selected frame does not detect a human figure, select one of the other two frames when group 2 is polled next time. After the frames in group 2 are detected, group 1 is detected, and the detection is repeated in this way. This ensures that one of the two inspections must be a fixed frame detection. This can solve the problem of people not being detected when they are still in the picture, and significantly reduce the missed detection rate.

[0042] In this embodiment, the BGM detection algorithm is used to detect the motion area in step 1, and the detected motion area is recorded. In step 1, the motion frame at the top of the video screen and the particularly small motion frame are filtered by a neural network algorithm.

[0043] In this embodiment, the alarm in step 3 and step 4 adopts one or more of sound alarm, light alarm and vibration alarm. In step 3, the detection of the human figure is marked by data processing. When the detection is cyclic, the detection is stopped until the human figure is detected and the alarm is sounded.

[0044] The humanoid filtering algorithm of the humanoid intelligent camera in the security monitoring field can significantly improve the rate of detecting humanoids by taking turns to inspect group one and group two using a poor detection method. At the same time, it ensures that one of every two inspections is a fixed frame detection. This can solve the problem of not being detected when a person is stationary in the picture, significantly reduce the missed alarm rate, and effectively reduce the time consumption of detection. If a humanoid appears at the top of the video screen, it means that the current distance between the humanoid and the camera lens is more than 10 meters, so the detection distance is finally set to 10 meters as the acceptance standard. On the basis of the humanoid algorithm, the humanoid detection results at the top of the video screen and the very small BGM results are filtered out, thereby reducing the false alarm rate of humanoids and increasing the detection distance to 10 meters, effectively improving the detection efficiency.

[0045] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A humanoid filtering algorithm for a humanoid intelligent camera in the field of security monitoring, characterized by: The humanoid filtering algorithm comprises the following steps: Step 1: Preparation for human figure filtering detection, obtain the results of the BGM motion area, and filter out the motion frame at the top of the video screen and particularly small motion frames; Step 2: Calibrate and process the detection frame. The current detection screen width and height are 640*368. Define 4 detection frames by yourself. The x, y coordinates and width and height are: Full frame: x=0, y=20, w=640, h=320; Middle box: x=20, y=100, w=600, h=120; Lower left frame: x=0, y=168, w=600, h=200; Lower right frame: x=40, y=168, w=600, h=200; The BGM motion frame and the full-frame frame are set as group one, and the middle frame, lower left frame and lower right frame are set as group two for cross detection, and group one and group two are sent for inspection in turn; Step 3: Detect group 1. First, detect all the motion frames of the BGM in group 1. If one BGM frame detects a human figure, the system will exit and alarm. Otherwise, if the BGM frame does not detect a human figure, the system will detect the full-frame frame once. If it detects a human figure, the system will exit and alarm. Step 4: Detect group 2. After group 1 is detected, group 2 is detected. Select one frame from the middle frame, lower left frame and lower right frame of group 2 for detection. If a human figure is detected, make a mark, exit and alarm. The next time group 2 is polled, select the frame for detection. If no human figure is detected after two consecutive pollings, switch to other frames for detection. If the selected frame does not detect a human figure, select one of the other two frames when group 2 is polled next time. After the frames in group 2 are detected, group 1 is detected, and the detection is cyclic.

2. The humanoid filtering algorithm for the humanoid intelligent camera in the field of security monitoring according to claim 1 is characterized by: In step 1, a BGM detection algorithm is used to detect the motion area, and the detected motion area is recorded.

3. The humanoid filtering algorithm for the humanoid intelligent camera in the field of security monitoring according to claim 2 is characterized in that: In step 1, a neural network algorithm is used to filter the motion frame at the top of the video screen and particularly small motion frames.

4. The humanoid filtering algorithm for the humanoid intelligent camera in the field of security monitoring according to claim 1 is characterized by: Step 2: In the process of sending Group 1 and Group 2 for inspection in turn, a cyclical manner is adopted for the inspection.

5. The humanoid filtering algorithm for the humanoid intelligent camera in the field of security monitoring according to claim 1 is characterized by: The alarm in step 3 and step 4 uses one or more of sound alarm, light alarm and vibration alarm.

6. The humanoid filtering algorithm for the humanoid intelligent camera in the field of security monitoring according to claim 1 is characterized by: In step 3, one box is selected from the middle box, the lower left box and the lower right box of group 2 for detection, and the selection is performed randomly.

7. The humanoid filtering algorithm for the humanoid intelligent camera in the field of security monitoring according to claim 1 is characterized by: In step three, the detected human figure is marked by using a data processing method.

8. The humanoid filtering algorithm for the humanoid intelligent camera in the field of security monitoring according to claim 1 is characterized by: In step 3, a loop detection is performed until a human figure is detected and an alarm is sounded, and then the detection is stopped.

Images