A vision-based vehicle detection, tracking and early warning method

Abstract

The invention provides a vision-based vehicle detection, tracking and early warning method. The method comprises the steps of collecting the image and calibrating the road vanishing line; defining andgraying the vehicle detection area, and then performing gray stretching on the vehicle detection area image according to the illumination intensity classification of the collected image; constructingthe training sample images and labeling the images as positive sample images and negative sample images manually; extracting the haar feature and LBP feature of the positive sample images and negative sample images to train an Adaboost cascade classifier; dividing the vehicle detection area into different domains, and detecting the vehicle by the Adaboost cascade classifier after training, and judging the vehicle twice according to the illumination intensity. When the vehicle is detected, the KCF target tracking method is used to track the vehicle. The vehicle is tracked, and the distance between the vehicle and the vehicle is calculated by the distance estimation method based on the position, and the collision time is calculated according to the vehicle speed and the distance between thevehicle and the vehicle to give the early warning. The method reduces the computational complexity and improves the vehicle detection accuracy.

Description

technical field [0001] The invention relates to intelligent transportation, automotive electronics and visual vehicle detection and tracking collision early warning methods, and relates to a vision-based vehicle detection and tracking early warning method. Background technique [0002] With the continuous growth of car sales, road traffic accidents are also increasing. On December 19, 2017, the State Administration of Work Safety and the Ministry of Transport jointly issued an announcement stating that although road traffic accidents in my country have declined significantly in recent years, they are still high. In 2016 A total of 8.643 million traffic accident reports were received, a year-on-year increase of 659,000 cases, an increase of 2.9% compared with the same period, resulting in a total of 63,093 deaths and 226,430 injuries, and the death rate per 10,000 vehicles was as high as 2.14. March 7, 2017, Transportation The Ministry of Transport issued the "Safety Technical ...

AI Technical Summary

Problems solved by technology

At present, vision-based vehicle detection methods are roughly divided into the following five categories: The first category is template-based vehicle detection methods, which need to create a large number of templates and parameters, and need to observe and update templates in real time. The shape differences caused by different shapes, scales, and deformations during travel lead to endless templates, so template-based vehicle detection is not suitable for mobile sites

The second type is the method based on optical flow, which realizes the acquisition of vehicle motion parameters according to the pixel gray level changes of the image sequence, so as to obtain information such as the position of the vehicle. This method can only detect dynamic vehicle targets

The third type is a feature-based method, which uses vehicle bottom shadows, vehicle outlines, vehicle edges, rear corners, lights, etc. to detect vehicles based on these prior knowledge. This method has simple features and low detection rate

The fourth category is the traditional machine learning method, which has fast detection speed and high accuracy, but the generalization ability of the model is poor

The fifth category is the target detection method based on deep learning. This method has strong recognition ability and high accuracy, but has a large amount of calculation and requires high hardware requirements.

Images