Rapid vehicle detection and classification method based on artificial intelligence

Abstract

The invention discloses a rapid vehicle detection and classification method based on artificial intelligence, which comprises the following steps of: 1, dividing a video stream into continuous single-frame images through decoding and inputting of the video stream of a local camera or a network camera; step 2, detecting vehicles in the single-frame image obtained in the step 1 by using gray-scale texture characteristics, and roughly selecting an area which may be a vehicle; step 3, sending the region of interest of the picture obtained in the step 2 into a trained convolutional neural network for classification and identification to obtain the specific classification of the vehicle; step 4, tracking the vehicles identified in the step 3 by using Kalman filtering, and outputting classification information of the vehicles when the same vehicle is detected in three continuous frames of single-frame images; and 5, carrying out statistics on traffic flow and vehicle type data information according to the information output in the step 4. According to the method, the vehicle detection and classification speed of the video stream is increased, the requirement for hardware is not high, andpractical application and popularization of the algorithm are facilitated.

Description

technical field [0001] The invention relates to artificial intelligence and image recognition, in particular to a rapid vehicle classification method based on artificial intelligence. Background technique [0002] Deep learning is a popular technical field this year. Among them, there are many object recognition and classification methods in the field of computer vision. Among them, the most common and effective object classification network is convolutional neural network (CNN). Most of the algorithms are also extended based on CNN. The more mature ones include R-CNN, SPP-Net, Fast RCNN, etc. These algorithms have basically reached the recognition level of human eyes in the recognition and classification of limited types of objects. However, most of these algorithms stay in the research stage and are not widely used in actual engineering projects, mainly due to the following reasons: First, the training and operation of CNN itself needs to consume a lot of hardware resource...

AI Technical Summary

Problems solved by technology

However, most of these algorithms stay in the research stage and are not widely used in actual engineering projects, mainly due to the following reasons: First, the training and operation of CNN itself needs to consume a lot of hardware resources, and the improvement of recognition accuracy is often accompanied by The most important thing is that the deepening of the network and the expansion of the input data further increase the burden on the hardware; when the requirement is extended to object recognition, it is essentially to select 1000-2000 regions from the image to be recognized and perform CNN classification. A good object recognition and classification model often requires a powerful GPU for calculation, and the hardware cost is very high when processing high data streams such as video, so it is difficult to be extended to practical engineering applications

Although there are some hardware computing units for deep learning algorithms, such as TPU, etc., there are still problems such as not open source, less supply, and difficult development.

Images