Target detection sliding window scanning method based on scene geometric constraint

Abstract

The invention discloses a target detection sliding window scanning method based on scene geometric constraint. An existing total graph sliding window exhaustive search method based on a monocular camera has the problem that the real-time performance of a detection system can not be achieved caused by low search efficiency. The invention discloses the target detection sliding window scanning method based on the scene geometric constraint. According to the method, by using eight control points and a direct linear transformation algorithm, the homography matrix of a pre-calibrated vehicle-borne camera and the projection vector of a ground plane normal vector on an image are calculated, the upper boarder and the lower boarder of a sliding window scanning constraint area are calculated according to a constraint equation deduced from image imaging geometric constraint relation, and the image characteristics pyramid construction and sliding window scanning detection and classification are carried out only in the corresponding image constraint area. The scanning efficiency is high, and since only one camera is needed, a problem of high hardware cost of a multi-stereo detector is solved.

Description

technical field [0001] The invention relates to a high-efficiency sliding window scanning method in target detection of vehicle-mounted video images, in particular to a video image sliding window scanning method based on scene geometric constraints. Background technique [0002] With the rapid development of transportation, traffic safety is becoming more and more important. Target detection and early warning of vehicles and pedestrians in advanced driver assistance systems is an important method to reduce vehicle traffic accidents. Image sliding window scanning search and detection is a key part of vehicle image target detection technology. An important factor affecting the efficiency and real-time performance of image target detection. [0003] Vision-based image target detection can be divided into two categories according to the different detectors used: the full-image sliding window exhaustive search based on monocular cameras and the sliding window search based on mul...

AI Technical Summary

Problems solved by technology

This method uses eight control points and direct linear transformation algorithm to calculate the homography matrix of the pre-calibrated vehicle camera and the normal vector of the ground plane in the image projection vector, and calculates the sliding window scan according to the constraint equation derived from the geometric constraint relationship of image imaging. The upper boundary and the lower boundary of the constrained area only perform image feature pyramid construction and sliding window scanning detection classification in the corresponding image constrained area, which not only has high scanning efficiency, but also solves the problem of multi-eye detector hardware because only one camera is needed. high cost problem

Images