summary
Researchers Wang Cong, Liu Mingguang, and others from the School of Electrical Engineering at Beijing Jiaotong University pointed out in an article published in the 9th issue of the journal "Electrical Technology" in 2018 that with the continuous improvement of the automation level of power systems, the application of intelligent video surveillance in power systems is becoming more and more common.
This paper reviews dynamic target detection and recognition algorithms used in intelligent video surveillance systems. It introduces video moving target detection algorithms, including inter-frame difference, background difference, and optical flow methods, as well as traditional template-based recognition methods and statistical learning-based recognition methods. The various algorithms are compared, and their applicable environments, advantages, and disadvantages are discussed.
In recent years, video surveillance technology has been increasingly applied to power systems. Installing video surveillance systems in unattended substations and dangerous areas where personnel are prohibited from access can effectively eliminate safety hazards, provide timely warnings, and prevent accidents. Traditional video surveillance systems often rely on staff on 24/7 for manual alerts. This approach not only fails to reduce accidents through alarm systems but also represents a significant waste of human and material resources.
Today, with the continuous advancement of computer vision research, intelligent video surveillance technology is gradually replacing traditional video surveillance methods. Intelligent video surveillance systems can analyze video images using image processing and computer vision methods without human intervention to determine the real-time status of the monitored location. When anomalies occur, they can promptly report to staff, prompting them to take appropriate measures, thereby achieving the functions of prevention, early warning, and proactive monitoring.
Currently, the most commonly used intelligent video surveillance algorithms are primarily based on moving target detection algorithms. These algorithms process the acquired video information frame by frame, issuing an alarm signal when an intrusion is detected within the monitored area. However, these algorithms cannot distinguish between moving objects and have high requirements for the application environment, making them unsuitable for complex environments. If advanced moving target detection algorithms are used to detect moving objects, and pattern recognition and machine learning methods are employed to identify them, intelligent video surveillance technology can achieve greater adaptability and accuracy.
1. Moving target detection algorithm (omitted)
Currently, the main video moving target detection algorithms that can be implemented and widely used include: inter-frame difference method, background difference method, optical flow method, and some advanced fusion algorithms.
2. Overview of Moving Target Recognition Algorithms (omitted)
After a moving target detection algorithm detects a moving area, different moving areas may correspond to different moving objects. The accuracy of alarms from intelligent video surveillance systems will be affected by the ability to correctly identify abnormal moving objects.
Meanwhile, target recognition technology can also be applied to automatic inspection systems for power lines to identify whether equipment is damaged. Currently, there are many types of moving target recognition algorithms, which can be broadly divided into traditional template-based object recognition methods and statistical learning-based object recognition methods.
3. Algorithm Comparison and Future Prospects
3.1 Algorithm Comparison
In summary, among basic moving target detection techniques, inter-frame difference and background difference methods, as fundamental moving target detection algorithms, possess advantages such as simple algorithms, low equipment requirements, and fast processing speed. However, they also have disadvantages such as poor adaptability and sensitivity to background changes such as illumination. Optical flow methods offer higher detection accuracy and can resolve the problem of target occlusion and overlap.
However, optical flow is computationally complex and requires a huge amount of data. Unless supported by specialized hardware, it is difficult to meet the requirements of real-time video processing. Furthermore, the limitations of the assumptions used in its computation make optical flow quite sensitive to noise. In future algorithm development, fusion algorithms combining various methods and more advanced background modeling methods will undoubtedly become research hotspots.
In object recognition algorithms, traditional template-based object recognition methods can be viewed as "deductive reasoning," where a human inputs the explicit features of an object, and the computer completes the recognition when it detects objects with high similarity to these features. This approach works well when the features of moving objects are clear and does not require a large number of original samples to support the algorithm. However, this algorithm fails when the features of the object to be recognized are not obvious or are not easily represented in machine language.
Object recognition methods based on statistical learning can be viewed as "inductive methods." During the recognition process, no rules need to be input into the computer; only a large number of representative samples from the real world are required. The algorithm can then autonomously extract specific rules to complete the recognition. This type of algorithm can extract high-level, abstract features of the object being tested and has good adaptability and accuracy in practice. However, its drawbacks include the need for a large amount of training data and computation, placing high demands on the equipment.
3.2 Future Outlook
In recent years, intelligent video surveillance technology has developed rapidly and its application in smart grids has increased. However, it still suffers from drawbacks such as high false alarm rates, narrow application scope, inability to fully intelligently identify corresponding faults, and the need for human intervention. From a technical perspective, intelligent video surveillance systems will inevitably develop in the following directions:
1) Better adaptability. Only by improving the adaptability of the algorithm can intelligent video surveillance better adapt to complex and ever-changing environments.
2) It can identify a wider variety of anomalies with higher accuracy. Current pattern recognition methods can achieve high accuracy, but deep learning-based methods, due to the need for a large number of positive and negative samples, have relatively low accuracy. Improving the accuracy of these methods will be a future research direction for intelligent video surveillance technology.
3) Intelligent video surveillance devices will inevitably develop towards diversification of product forms. Currently, the most commonly used method is wired transmission with externally mounted cameras. In order to adapt to diverse working environments, intelligent video surveillance devices will also develop towards various terminal forms such as portable handheld, vehicle-mounted, and remote-controlled, as well as wireless video transmission.
in conclusion
Intelligent video surveillance technology is a research area with significant practical application value. Furthermore, with the continuous improvement of automation in my country's power system, intelligent video surveillance technology is increasingly being used in intrusion detection, equipment status monitoring, and security alarms. Currently, moving target detection technology based on video information has gradually matured and is being applied more and more in practice.
However, the technology for automatic identification of moving targets is still in the exploratory and research stage. Much work remains to be done to truly design an intelligent video detection and identification system suitable for power systems. It is believed that with the continued advancement of related research, intelligent video surveillance technology will undoubtedly shine brightly in power systems.
