Vehicle identification technology is one of the most important aspects of intelligent transportation systems. Vehicle identification, image preprocessing, feature extraction and other technologies are used to identify vehicle grades. Today, license plate recognition is widely used in high-speed toll stations, parking lots, etc., and China's camera recognition technology is at the top of the world, but due to the impact of scanning technology on light, dirty cars, license plate damage, etc., the comprehensive accuracy of hardware recognition It can only stay at around 95%, which also leads to no sense of payment, unattended as a paper.

First, the license plate color recognition

In the aspect of vehicle color recognition, it basically overcomes a series of problems such as color instability and overexposure caused by changes in illumination conditions and camera hardware errors, thus solving the recognition error caused by image color change.

Second, vehicle search

In the aspect of vehicle retrieval, the picture of the vehicle may be overexposed or underexposed in different scenes, or the scale of the vehicle changes greatly, which causes the characteristics extracted by the traditional method to change, so the retrieval rate is unstable. In the face recognition project, the face changes due to factors such as light, posture and expression. Many applications are fixed scenes and fixed poses. After using the deep learning algorithm, not only the face recognition rate of fixed scenes is increased from 89% to 99%, and there is some relaxation in posture and light.

Third, license plate recognition

In terms of vehicle identification, deep learning-based vehicle identification technology extends the range of features from a simple license plate or logo to the entire body. The vehicle's lights, grille, window and so on are all important features of the vehicle. The introduction of these features not only greatly improves the accuracy of vehicle identification, but also adapts to problems such as interference and occlusion. The categories are also more detailed, not only identifying the brand of the vehicle, but also identifying detailed categories such as sub-brands, models, and models of the vehicle. The retrieval of the designated vehicle in the video image data can be performed by means of description information such as license plate, brand, model, color, etc., and can also be performed by local features such as vehicle pictures or annual inspection labels and ornaments.

At present, the vehicle bayonet system in many cities in China expands the vehicle identification function based on the existing system, also known as the vehicle secondary analysis system, which can basically identify more than 2,000 types of vehicles that are refined to the annual model. Based on the expansion of a number of practical business applications such as "fake / deck analysis". Especially for "fake cards", "deck cards", "no seat belts in the car", "driving calls" and other illegal activities that require manual screening, these intelligent transportation systems not only do more with less, but also greatly reduce manual investment. Increase work efficiency.