Analysis of the reason why the bayonet recognition accuracy is too low

1. External road environmental impact

Due to the different installation and installation locations of different traffic bayonet, the degree of influence on the external road environment is also different. Some bayonets are not affected by the lack of light in the surrounding environment, the illumination is too strong, the rainy lens is wet, the heavy trucks are vibrated by equipment, the trees are blocked, etc., resulting in insufficient or too large pixel points in the number plate of the passing picture captured by the bayonet. It will affect the recognition accuracy of the number plate. It is based on the training experience of a large number of plate samples in the monitoring industry. Under normal circumstances, the pixel area size of the card area is between 100 and 140. It is difficult to guarantee the number outside the range. Card recognition accuracy. For example, under strong and strong backlight conditions, the video image is easy to be too dark or too violent, which causes the image brightness distortion to affect the normal processing of the number plate recognition algorithm; in the windy weather, the L-bar of the mounting bayonet is easily stressed to generate the arm swinging The constant change of the scene makes the algorithm unable to effectively locate the vehicle and the number plate model, thus affecting the normal recognition.

2. Device hardware performance issues

Since the traffic bayonet device has been exposed to the outdoors, it cannot be affected by the climate and the environment. Some of the traffic bayonet cameras have a downward trend with the use of time, which leads to a decline in the accuracy of passing picture recognition. In addition, from the picture of some bayonet, the vehicle number plate is clear and complete, but there are still cases of digital letter confusion. The camera equipment of some manufacturers also has insufficient hardware performance configuration and insufficient algorithm analysis accuracy. Taking Zhejiang Dahua bayonet equipment as an example, three generations of bayonet cameras have appeared in the past five years. The first generation is a common series of 5 million to 7 million pixels listed before 2016; the second generation is 700 listed between 2016 and 2017. The megapixel "E" series; the third generation is the 7-900-900 pixel "Hui" series that was launched after 2018, and the later products have greatly improved in the deep learning and optimization of image analysis algorithms.

3. Special vehicle impact

In 2017, new energy vehicles were put into use. Due to the different non-new energy vehicles, some of the early construction of bayonet cameras could not support the identification of new models, which may lead to identification errors. According to feedback from camera manufacturers such as Hikvision and Zhejiang Dahua, the front-end equipment that was manufactured before 2016 cannot support the effective identification of new energy license plates. In addition, there are some medium and heavy truck drivers who are monitored to capture or punish in order to avoid vehicles. Some technical means are used to interfere with the normal capture of the bayonet, thus affecting the accuracy of the card image recognition. For example, the high beam of the modified vehicle is subjected to strong light interference when the vehicle passes the position of the bayonet, so that the bayonet cannot be recognized normally, or the card is blocked or deliberately defaced to avoid the normal capture and evidence collection of the bayonet.

4. Equipment operation and maintenance is not in place

In recent years, the investment in the construction of traffic bayonet has continued to increase, forming a large-scale situation. However, due to the concept and mechanism of equipment operation and maintenance management, some bayonet equipments are leaking or not in place. Conditions, such as camera installation angle deviation, inaccurate focus, lens not regularly wiped, algorithm program not updated in time, fill lamp damage or brightness drop, etc., also affect the bayonet image recognition accuracy to varying degrees.

5. Standard requirements are too high

Whether the card slot number recognition accuracy meets the actual application needs, the direct or effective comparison basis is the corresponding technical standard. However, according to the operation of the bayonet in recent years, and the tracking research of some bayonet equipment manufacturers, the existing national standard has been set too high for the identification rate of the bayonet number plate (such as more than 95% during the day and more than 90% at night), or It can be regarded as a “laboratory standard”. It is not very grounded. The standardization does not take into account the differences between different environments, different models, equipments of different use periods, etc., and the identification accuracy of the number plate recognition is caused. (Author: Tao Yuan Shao Xiaobo latent)