Vehicle detection method based on vehicle camera in night vision environment

Abstract

A vehicle-mounted camera-based vehicle detection method in a night vision environment relates to a method for detecting vehicles ahead. It first performs exposure control and polarized light elimination on the vehicle-mounted camera that detects and tracks vehicle lights, and then uses a combination of The region growing method of canny edge detection and the color image threshold segmentation method based on the Mahalanobis distance are used to segment the headlights and taillights; on this basis, the headlights are paired, and the calculation accuracy of the correlation coefficient is improved by preliminary filtering and projection distortion compensation; finally, the The target is tracked and confirmed based on Kalman filter and state machine conversion. The invention has the advantages of reducing complex background interference, reducing reflection intensity on wet roadsides, detecting taillights, maintaining the appearance shape of headlights, improving calculation accuracy of correlation coefficients, simple calculation and solution process, speeding up detection speed and maintaining accurate vehicle detection It can realize accurate detection and stable tracking of the vehicle in front in the night vision environment.

Description

technical field [0001] The invention relates to a method for detecting a vehicle in front, in particular to a method for detecting a vehicle in front based on a vehicle-mounted camera in a night vision environment. Background technique [0002] Other vehicles on the road are information that the driver must pay attention to during driving. The assisted driving system that automatically detects the position of other vehicles has a wide range of uses, such as: collision warning, blind spot monitoring, automatic cruise, etc. In-vehicle visible light cameras are an inexpensive sensing device widely used for such functions. But at night, the visual information in visible light video is very limited. Many background information such as road signs, lane lines, horizon lines and object information such as body structure and vehicle color are almost completely unavailable. In the dark night environment, the most prominent vehicle information is the vehicle's headlights and taillig...

AI Technical Summary

Problems solved by technology

However, the method in this paper does not calculate the symmetry-related information of the lights. It is simply considered that a pair of lights on the same horizontal axis belong to the same target vehicle. At the same time, as pointed out at the end of the paper, the method in this paper is vulnerable to The colored road signs or billboards on both sides of the road interfere. In addition, the rain on the road will also affect the detection effect of the method in this paper.

Images