Method for accurately positioning moving object based on deep learning

Abstract

The invention discloses a method for accurately positioning a moving object based on deep learning. The method comprises the following steps of: a, acquiring a to-be-detected video sequence and a corresponding depth map; B, adopting darknet-yolo-v3 to detect a moving target in the video sequence and marking an identification box; C, combining with depth-of-field information in the depth map, searching a contour in the depth map by adopting a correlation function of Opencv, and drawing a rectangular boundary surrounding the contour to obtain a rectangle of an interested area; D, calculating thearea of the identification frame, the central point of the identification frame, the rectangular area and the rectangular central point; and e, matching the area of the identification frame with thecentral point of the identification frame, the rectangular area and the rectangular central point, and when the two are matched within a preset threshold range, determining that the position of the identification frame is the position of the moving target. According to the method, the cavity phenomenon can be avoided, the real-time performance is high, and the recognition accuracy is high.

Description

technical field [0001] The invention relates to a moving object positioning method, in particular to a method for precise positioning of a moving object based on deep learning. Background technique [0002] Moving object detection refers to the process of subtracting redundant information in time and space in video through computer vision, and effectively extracting objects that change in spatial position. Research in this direction has always been an important research topic in the field of computer vision. When detecting moving targets in video streams, precise positioning of moving objects has become the most challenging research direction in the field of computer vision research, and it involves many cutting-edge disciplines, such as: deep learning, image processing, model Identification, etc., combined with these disciplines has become a research hotspot. [0003] In many scenarios, such as in the security monitoring system of important large places such as high-speed...

AI Technical Summary

Problems solved by technology

This method is more delicate, but the amount of calculation is large, resulting in a decrease in real-time performance, which cannot meet the needs of some systems with high real-time requirements

At present, some researchers have proposed the optical flow method, but most of the optical flow methods are quite complicated to calculate and have poor anti-noise performance. If there is no special hardware device, they cannot be applied to real-time processing of full-frame video streams.

Images