Pedestrian multi-target tracking method based on CenterNet and depth incidence matrix
A multi-target tracking and correlation matrix technology, applied in the field of pedestrian multi-target tracking based on CenterNet and deep correlation matrix, can solve the problems of complex framework and low efficiency, and achieve the effect of ensuring high precision, high efficiency and reducing network parameters
Active Publication Date: 2021-03-16
余姚市浙江大学机器人研究中心 +1
View PDF6 Cites 2 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0008] Aiming at the problems of complex frame and low efficiency of the existing pedestrian multi-target tracking algorithm, the present invention provides a pedestrian multi-target tracking method based on CenterNet and depth correlation matrix
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment
[0067] Training phase:
[0068] (1) The collected video is framed into a picture sequence of 65,000 pictures in total, and the pedestrians in each frame of picture I are marked with a rectangular frame, and each rectangular frame is marked as
[0069] (2) if figure 1 As shown, the marked adjacent data pair I t with I t+1 Input the CenterNet network to extract pedestrian features respectively, and network parameters are shared;
[0070] (3) CenterNet detection network predicts picture I t There are M pedestrians in total. I t+1 There are N pedestrians in total. Each pedestrian position uses a rectangular box D d =(x, y, w, h) represents, wherein, x, y represent the coordinates of the upper left vertex of the rectangular frame. w, h represent the length and width of the rectangular frame;
[0071] (4) Calculate the L2 loss L between the network detection result and the labeled true value d =||D d -D t || 2 , and calculate the center (Cx, Cy) of each pedestrian fra...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a pedestrian multi-target tracking method based on a CenterNet and a depth incidence matrix, and the method comprises the steps: extracting pedestrian features in an input image through employing the CenterNet, detecting the position of a pedestrian in the input image in a rectangular frame mode, then calculating the center of each rectangular frame, and extracting the features corresponding to the center in each stage of a CenterNet network; splicing the extracted features to construct a feature association matrix, and performing compression and association matching onthe extracted features by adopting a deep affinity network DAN, thereby realizing pedestrian multi-target tracking in a complex scene. The pedestrian multi-target tracking method is a pedestrian multi-target tracking technology with high performance and high efficiency, an innovative mode of detection and Id association matching shared feature extraction network and joint training is adopted, efficiency is reduced, and meanwhile high tracking precision is guaranteed.
Description
technical field [0001] The invention relates to the field of computer vision, in particular to a pedestrian multi-target tracking method based on CenterNet and depth correlation matrix. Background technique [0002] Pedestrian object tracking is an important topic in the field of computer vision. It integrates knowledge in various fields such as artificial intelligence, image processing, pattern recognition, and automatic control, and has important applications in the fields of intelligent video surveillance, human-computer interaction, robot visual navigation, target recognition, traffic detection, and motion analysis. In addition to the above civilian applications, moving target detection and tracking technology based on image sequences also has broad application prospects in the military field, especially in guidance and navigation. [0003] Pedestrian object tracking is difficult and lacks high-quality training data, so although object tracking has been extensively stud...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/00G06K9/62G06N3/04
CPCG06V40/25G06V40/10G06N3/045G06F18/214
Inventor 王文靖段志钊张建明王志坚
Owner 余姚市浙江大学机器人研究中心