Vision-based vehicle door opening collision warning device and method

Abstract

The invention provides a vision-based car door opening collision warning device and method, including four fish-eye cameras and an industrial computer, using four fish-eye cameras installed on the head, tail, left and right of the vehicle, using fish-eye The angle of view of the eye camera is close to 180 degrees, and it completes the observation of the car without dead angles, and forms a surround-view platform with the on-board industrial computer; the images of the four fields of view obtained by the surround-view platform have fisheye distortion, and distortion correction is required; The images after correction of field of view distortion are detected based on STHOL features respectively. The invention has low requirements on hardware, good maintainability, good portability and good expandability.

Description

technical field [0001] The present invention relates to an auxiliary driving technology in the traffic field, in particular to a vision-based car door opening collision warning device and method. Background technique [0002] The research on intelligent vehicles is one of the current research hotspots. In the study of smart cars, the driving assistance system is one of the core systems. Among the various behaviors of using a vehicle, the source of the danger of many behaviors is the visual blind spot that exists when the people in the vehicle perform the behavior. The behavior of parking and opening the door is one of the above-mentioned dangerous behaviors, because collision accidents caused by occupants ignoring moving objects in their visual blind spots when parking and opening the door occur from time to time. Therefore it is very necessary to develop a kind of advanced principle, simple, high accuracy and moderate price vehicle door collision warning device. [0003]...

AI Technical Summary

Problems solved by technology

However, most of the existing detection devices using machine vision can only realize the visual detection of some areas around the observation point, and there are few devices that can realize the global detection of blind spots around the car body, and most of the algorithms used by the devices are optical flow methods. These traditional algorithms have their own shortcomings. For example, the optical flow method has a huge amount of calculation and low calculation efficiency; the frame difference method will produce a large error when the moving speed of the object is too high or too low; The background subtraction method takes a long time to model the background, and the initialization time is long, etc.

Images