Moving object detection method of mobile camera under high parallax

Abstract

The invention proposes a moving camera moving target detection method under strong parallax. Use the rotation matrix R and the translation matrix t to describe the camera motion between two frames, and obtain the relationship between the points on the second frame image and the world coordinate system, so as to obtain the depth constraint equation; the coordinates of all matching points on the second frame image and The corresponding depth information is brought into the depth constraint equation to obtain the optimal solution of the Rt matrix of the depth constraint equation; the image point coordinates calculated by the depth constraint equation are multiplied by the corresponding depth information; The results are respectively used as two three-dimensional coordinate points, the distance between the two three-dimensional coordinate points is calculated, and the distance is normalized. If the normalized distance is greater than the set threshold, it is judged that the point corresponding to the distance is a moving target. Otherwise, the point corresponding to this distance is the background. The invention can solve the problem of parallax in three-dimensional scenes under mobile platforms such as vehicle-mounted handhelds, and can detect moving objects in real time and efficiently.

Description

technical field [0001] The invention belongs to the field of digital image processing and pattern recognition, in particular to the field of target recognition and tracking based on image processing, and specifically relates to a moving target detection method of a mobile camera under strong parallax. Background technique [0002] Automatic moving target detection technology is the key technology in the target search and tracking system, which provides preliminary information for subsequent track association, target recognition, tracking and other technologies. In the traditional moving object detection system, the camera is usually stationary, and it is widely used in video surveillance, traffic detection and other systems. However, in mobile platforms such as vehicles and handhelds, due to the movement of the camera, the 3D scene will produce parallax. At this time, it is necessary to effectively estimate the scene parallax. [0003] Traditional methods generally use the ...

AI Technical Summary

Problems solved by technology

Among them, the homography matrix uses the mapping relationship between matching point pairs on the same plane, which is especially suitable for situations where the scene parallax can almost be ignored for drone aerial photography; the basic matrix uses the emitter geometric relationship between image matching point pairs to detect moving objects ; The multi-viewpoint image used by the trifocal tensor is increased from two views to three views, and the mathematical quantity representing the intrinsic projective geometric relationship between images that does not depend on the scene structure is also changed from the basic matrix to the trifocal tensor, but the calculation more complicated

[0004] However, the above traditional methods require a reference calibration object with known geometric information, and the calibration method based on active vision needs to use accurate pan / tilt and other hardware platforms to provide the orientation information of the camera at any time, which has extremely high hardware requirements and is not suitable for portable cameras; Self-calibration methods such as basic matrix and three-view tensor, although not demanding on calibration scenes and calibration instruments, often have a large workload, poor robustness, and low accuracy

Images