Method for real-time detection of road vehicle based on deep learning SSD frame

Abstract

The invention discloses a method for real-time detection of a road vehicle based on a deep learning SSD frame and mainly solves a problem of low detection accuracy of the existing technology under the conditions of traffic congestion and complicated weather. An implementation scheme of the method comprises the following steps that 1.shooting videos of a plurality of running vehicles at a vital communication line and grabbing window information of the vehicle in each frame of image in the video in a manual labelling way; 2.taking VGG-16 in a classification network as a basic network, constructing an SSD300*300 detection frame, extracting features of some feature extraction layers to carry out connection as recognized feature vectors and inputting the feature vectors into a loss function; 3.training through a training sample and taking a trained detection model into the detection frame; and 4.setting a correlation threshold and detecting the vehicles in the test videos by the trained detection model. According to the method, the detection accuracy is greatly improved and an effect of real-time detection is realized, and the method can be used for vehicle detection under the complex scene.

Description

technical field [0001] The invention belongs to the technical field of image recognition, and mainly relates to a real-time detection method for road vehicles, which can be used for detection of running vehicles. Background technique [0002] With the continuous improvement of living standards in modern society, as an important means of transportation, the number of automobiles shows a trend of rapid growth, which brings great challenges to traffic supervision. Although the rapid increase in the number of cars has brought a lot of convenience to people's lives, it has also brought a series of traffic problems such as running red lights, traffic jams, speeding, and traffic accidents. The traditional road traffic management methods relying on manpower or basic transportation facilities can no longer meet the current development needs. In recent years, technologies such as computer technology, artificial intelligence and pattern recognition have been vigorously developed and wi...

AI Technical Summary

Problems solved by technology

The traditional road traffic management methods relying on manpower or basic transportation facilities can no longer meet the current development needs. In recent years, technologies such as computer technology, artificial intelligence and pattern recognition have been vigorously developed and widely used, and these technologies have also penetrated into traffic services. Industry, in this context, traffic monitoring technology based on deep learning convolutional neural network came into being

[0003] At present, in the research on car model recognition based on deep convolutional neural network, "Car Model Recognition Based on Deep Convolutional Neural Network" by Deng Liu of Southwest Jiaotong University, the research mainly distinguishes vehicle types such as cars, trucks and buses, but this The research is based on the small block of vehicle image as input, the classification and recognition of the combination of simple convolutional neural network and SVM, and does not include the research on the recognition of vehicle types in harsh environments such as rainy days, foggy days, snowy days and congestion.

This kind of small image block as input cannot combine background information well, and capturing the entire vehicle information makes the deep network capture a lot of redundant information that has nothing to do with classification and recognition, and the classification and recognition network cannot achieve end-to-end detection , not only the detection accuracy is not high, but also causes a waste of detection time, which cannot meet the requirements of real-time detection

Images