Animal fighting behavior real-time monitoring camera system based on edge calculation

Abstract

The invention discloses an animal fighting behavior real-time monitoring camera system based on edge computing, which adopts an edge computing technology and designs a simple algorithm based on YOLO deep learning aiming at the actual condition that the computing power of edge computing equipment is limited, so that fighting behavior recognition can be quickly carried out, and the real-time monitoring of the animal fighting behavior is realized. Therefore, the burden of the server for monitoring the camera system in real time is reduced, and the monitoring efficiency and the response speed are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of video surveillance, and in particular to the technical field of intelligent monitoring of fighting behavior. Background Art

[0002] Video surveillance technology is rapidly advancing in the field of target behavior recognition, with intelligent monitoring of fighting behavior attracting significant attention. In experimental monkey farms, real-time monitoring of fighting behavior is crucial for safeguarding the welfare of experimental animals, promoting vaccine development, and promoting public health. Given the critical role of experimental monkeys in scientific research, accurate monitoring of their behavioral patterns is crucial. However, traditional manual observation methods are time-consuming and inefficient, failing to meet the real-time and accuracy requirements. Therefore, the development of an efficient and intelligent system for identifying fighting behavior in experimental monkeys is urgent.

[0003] The state of the art related to the present invention:

[0004] The invention is titled "A method for detecting indoor fighting behavior in experimental monkeys based on target tracking" and the Chinese patent application number is 2024109653990. It introduces a method for detecting indoor fighting behavior in experimental monkeys based on target tracking. Although it can effectively identify fighting behavior, it relies on a central server to process a large amount of video data, ignoring the server load problem caused by the surge in data. Summary of the Invention

[0005] To solve the problems stated in the background technology section, the present invention introduces edge computing technology and designs a real-time monitoring camera system to directly process and analyze data on the camera side, reducing the server burden, improving monitoring efficiency and response speed, and thus adapting to the growing demand for video data in precision farming.

[0006] The present invention solves the technical problem as follows: a real-time monitoring camera system for animal fighting behavior based on edge computing, comprising one or more monitoring modules, a gateway, and a server, wherein the monitoring module is connected to the server via the gateway;

[0007] The present invention is characterized in that: the monitoring module includes a main control chip, a camera module, a sound pickup module and a transmission module. The main control chip receives the audio and video data input by the camera module and the sound pickup module, processes and identifies fighting behavior in real time, and then transmits the recognition results and related audio and video data to the server via the transmission module via the gateway. That is, the present invention adopts an edge computing method, which enables the monitoring module to directly process and analyze the captured audio and video data without transmitting all data to the server, reducing dependence on the server, greatly reducing data processing delays, and improving the real-time nature of fighting behavior monitoring;

[0008] In addition, in view of the limited computing power of edge computing devices, the present invention designs a fast algorithm for fighting monitoring, which improves monitoring efficiency while adapting to the computing power of edge computing devices. Specifically:

[0009] The present invention adopts a target monitoring algorithm based on YOLO deep learning to monitor animals in the video in real time, track the monitored animals, obtain the coordinates of each individual, and calculate the average displacement of multiple targets per unit time. When the average displacement exceeds a threshold and the sound volume also exceeds a threshold, it is judged that fighting behavior has occurred.

[0010] As a preference:

[0011] The monitoring module adopts H.265 high-efficiency video coding technology to reduce data transmission volume and improve transmission efficiency.

[0012] The monitoring module optimizes the YOLO model using techniques such as quantization and pruning, reducing model complexity, computational complexity, and memory usage, thereby improving operational efficiency. Quantization involves converting the model's high-precision data types to lower-precision data types, such as converting 32-bit floating-point data types to 8-bit integer data types. Pruning involves, for example, setting weights below a certain threshold to zero based on weight size.

[0013] The model update of the monitoring module adopts an asynchronous update strategy to reduce the delay and overhead of the model update.

[0014] The monitoring module also includes an alarm module, which is connected to the main control chip. When fighting behavior is identified, the main control chip transmits instructions to control the alarm module to issue an alarm for automatic intervention.

[0015] The main control chip also performs pre-processing operations such as denoising and enhancement on the raw video data it receives to improve the video quality, including the steps of using background subtraction and other technologies to remove static background and highlight animal movement information.

[0016] This invention uses edge computing and video analysis and recognition algorithms to quickly and accurately identify fighting behaviors. Compared with traditional technologies, it mainly solves the following problems:

[0017] 1) To address the slow and unstable video data transmission issues in traditional video surveillance systems, this invention uses edge computing technology to push computing tasks to the edge of the network, eliminating the need to transmit all data to the server as before. This can significantly reduce bandwidth usage and improve the real-time and reliability of data transmission.

[0018] 2) In response to the heavy burden and low speed problems caused by the central server processing large amounts of video data in the existing technology, the present invention achieves the dispersion of data processing pressure by integrating a fighting behavior analysis algorithm adapted to the computing power of edge devices, which can effectively shorten the data processing delay, support computing-intensive applications in resource-constrained environments, and realize real-time monitoring of fighting behavior. It provides solid technical support for the management and research of experimental animals and can also reduce the cost required to configure high-performance servers and large-capacity storage devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a system architecture diagram of a camera system for real-time monitoring of animal fighting behavior based on edge computing in a preferred embodiment of the present invention;

[0020] Figure 2 for Figure 1 A framework diagram of the monitoring module in the embodiment;

[0021] Figure 3 、 4 , 5 are three schematic diagrams of the behavior recognition algorithm of this embodiment. DETAILED DESCRIPTION

[0022] The present invention will be further described in detail below with reference to specific embodiments.

[0023] A real-time monitoring camera system for animal fighting behavior based on edge computing, the system architecture is as follows Figure 1 Shown, including:

[0024] Monitoring module, in this embodiment, is arranged as an edge device in the experimental monkey breeding room. Each experimental monkey breeding room can be equipped with one or more, where multiple means more than two. Figure 2 As shown, it includes a main control chip and a camera module, a sound pickup module, a transmission module, a lighting module, a speaker module, a power supply module and a transmission module respectively connected to the main control chip. The main control chip contains hardware CPU, GPU, RAM, and also includes a software recognition algorithm. The recognition algorithm processes and recognizes the fighting behavior of the experimental monkeys in real time based on the audio and video data input by the camera module and the sound pickup module. When fighting behavior is recognized, the main control chip transmits instructions to control the lighting module to emit strong light and the speaker module to simulate human voice to intervene. At the same time, the recognition results and related audio and video data are transmitted to the gateway in real time through the transmission module, eliminating the need to upload all audio and video data. The power supply module supplies power to the main control chip and other modules.

[0025] Gateway: used to receive audio, video and result data from each breeding room monitoring module, and transmit the audio and video data to private cloud storage.

[0026] Private cloud storage: used to receive audio, video and result data from the gateway and transmit the result data to the private cloud service.

[0027] Private cloud service: used to receive result data from private cloud storage, transmit the results to mobile clients and PCs in real time, and transmit the result data to public cloud services at the same time.

[0028] Mobile client and PC client: used to receive result data from private cloud services, remind administrators through pop-up windows, and issue abnormal alarms and early warnings.

[0029] Public cloud service: used to receive result data from private cloud services and perform data visualization, such as summarizing and organizing recognition results, and displaying information such as the experimental monkey's fighting behavior and movement trajectory on the user interface through the visualization tool ECharts.

[0030] Edge computing devices have limited computing power. To improve monitoring efficiency, this embodiment uses the following simple recognition algorithm to quickly and accurately identify fighting behavior:

[0031] It uses a target detection algorithm based on YOLO deep learning to monitor animals in real time in the video, track the detected animals, obtain the coordinates of each individual, and calculate the average displacement of multiple targets per unit time. When the average displacement exceeds a threshold and the sound volume also exceeds a threshold, it is judged to be fighting.

[0032] 1) If Figure 3 As shown in the figure, there are n experimental monkeys in an experimental monkey breeding room as an example (n = 3 in the figure). Through the target tracking algorithm, the recognition frame is calibrated for each experimental monkey. The coordinates of the upper left corner of the recognition frame are (x, y), the width and height are (w, h), and (x i,t ,y i,t ,w i,t ,h i,t ) represents the recognition box of the i-th experimental monkey at time t, where i∈[1,n]. When the experimental monkey moves in the video, the recognition box moves with the experimental monkey.

[0033] 2) If Figure 4 As shown, the coordinates (x o,i,t ,y o,i,t ) represents the center point of the recognition box of the i-th experimental monkey at time t, where x o,i,t =0.5×[(x i,t +w i,t )+x i,t ], y o,i,t =0.5×[(y i,t +h i,t )+y i,t ]. After Δt time, the center point moves to (xo,i,t+Δt ,y o,i,t+Δt ).

[0034] 3) If Figure 5 As shown, after Δt time, the distance moved by the i-th experimental monkey from time t to time t+Δt is d i,t+Δt , whose value is

[0035] 4) Calculate the average distance traveled by all monkeys in the breeding room from time t to time t+Δt Its value is The average moving speed is Set the threshold v, when It is believed that there is a possibility of fighting among the monkeys in the breeding room based on visual judgment.

[0036] The monitoring module of this embodiment also performs the following operations:

[0037] After receiving the raw video data, the video data is first pre-processed by denoising, enhancement and other operations to improve the video quality. This includes using techniques such as background subtraction to remove static background and highlight the movement information of the experimental monkey.

[0038] Compression algorithms are used to reduce the amount of audio and video data transmitted and improve the utilization of network bandwidth.

[0039] Adopt H.265 high-efficiency video coding technology to reduce data transmission volume and improve transmission efficiency.

[0040] Optimizing the YOLO model using techniques such as quantization and pruning reduces model complexity, computational overhead, and memory usage, improving operational efficiency. Quantization involves converting the model's high-precision data types to low-precision data types, such as converting 32-bit floating-point data types to 8-bit integer data types. Pruning, for example, sets weights below a certain threshold to zero based on weight size.

[0041] An asynchronous update strategy is adopted to reduce the delay and overhead of its YOLO model update.

[0042] By adopting edge computing technology and developing recognition algorithms adapted to the computing capabilities of edge devices, the present invention enables edge devices to directly process and analyze captured audio and video data without transmitting all data to the cloud or data center, reducing dependence on central servers and system bandwidth usage, thereby greatly reducing data processing delays, improving the real-time nature of fighting behavior monitoring, and the real-time and reliability of system data transmission.

[0043] The system of the present invention can maintain good stability in large-scale application scenarios and has good scalability.

[0044] The present invention can avoid the high cost required for configuring high-performance servers and large-capacity storage devices, reduce system construction and operation and maintenance costs, and improve performance-price ratio.

[0045] The system of the present invention has powerful computing power and high stability, and can issue an early warning within 3 seconds. It is not only suitable for monitoring the fighting behavior of experimental monkeys, but can also be expanded to other areas such as behavior monitoring and safety monitoring of other animals, and has broad application prospects.

Claims

1. A real-time monitoring camera system for animal fighting behavior based on edge computing, comprising one or more monitoring modules, a gateway, and a server, wherein the monitoring module is connected to the server via the gateway; Its characteristics are: The monitoring module includes a main control chip, a camera module, a sound pickup module and a transmission module. The main control chip receives the audio and video data input by the camera module and the sound pickup module, processes and identifies fighting behavior in real time, and then transmits the recognition results and related audio and video data to the server through the transmission module via the gateway; The main control chip uses a target monitoring algorithm based on YOLO deep learning to monitor animals in the video in real time, track the monitored animals, obtain the coordinates of each individual, and calculate the average displacement of multiple targets per unit time. When the average displacement exceeds a threshold and the sound volume also exceeds a threshold, it is determined that fighting behavior is present.

2. The animal fighting behavior real-time monitoring camera system based on edge computing according to claim 1 is characterized in that: The monitoring module encodes the relevant audio and video data using H.265 high-efficiency video coding technology.

3. The animal fighting behavior real-time monitoring camera system based on edge computing according to claim 1 is characterized in that: The monitoring module optimizes its YOLO model using techniques including quantization and pruning.

4. The animal fighting behavior real-time monitoring camera system based on edge computing according to claim 1 is characterized in that: The YOLO model of the monitoring module is updated using an asynchronous update strategy.

5. The animal fighting behavior real-time monitoring camera system based on edge computing according to claim 1 is characterized in that: The monitoring module also includes an alarm module, which is connected to the main control chip. When fighting behavior is identified, the main control chip transmits instructions to control the alarm module to issue an alarm for automatic intervention.

6. The animal fighting behavior real-time monitoring camera system based on edge computing according to claim 1 is characterized in that: The main control chip also performs pre-processing operations including denoising and enhancement on the raw video data it receives, including the steps of removing static background by background subtraction and highlighting animal movement information.

Images