Front vehicle detection method

Abstract

The invention discloses a front vehicle detection method. A mutual relation among adjacent subwindows in a sliding window method is considered. During a sliding window detection process, according to a classifier output result, a sliding step length is adjusted, unnecessary calculating is reduced and real-time performance is increased. During an adaptive promotion learning classifier detection process, a distinguishing strategy is set so that weak classifiers of each layer can carry out distinguishing on characteristic type attributes of images to be detected in a current layer and calculating cost is reduced. During an adaptive promotion learning classifier training process, a training sample is changed, the samples which can be correctly classified are deleted, a new training sample is increased, a classification boundary is further approached and detection precision is increased.

Description

technical field [0001] The invention relates to a method for detecting a vehicle in front. Background technique [0002] The detection and tracking of multiple vehicles ahead is the main basis for the intelligent decision-making of the forward collision warning system. The forward collision warning system can monitor the distance and speed of the vehicle in front relative to the vehicle in real time. By calculating the relative reaction time, it can give an early warning when there is a potential rear-end collision risk, prompting the driver to reduce the speed and maintain a relative distance and a certain relative speed. Therefore, in the complex and changeable urban road and high-speed road environment, timely and accurate detection of the vehicle in front becomes the key to the performance guarantee of the forward collision warning system. [0003] In a complex and changeable traffic environment, various uncertain factors such as the shape, color, scale, background, par...

AI Technical Summary

Problems solved by technology

Comaschi, Stuijk, Basten, RASW-A run-timeadaptive sliding window to improve Viola-Jones object detection, DistributedSmart Cameras (ICDSC), 2013Seventh International Conference on, Palm Springs, CA, pp:1-6, IEEE] for each sub-image When the window is detected, since the target only exists in a few windows, most of the window content is other irrelevant background content. Therefore, when continuing to use the adaptive lifting method to pass all the features in each window through all weak classifiers for classification and detection , the amount of calculation is particularly large and it is not necessary

Images