Long-time target tracking method based on depth detection
What is Al technical title?
Al technical title is built by PatSnap Al team. It summarizes the technical point description of the patent document.
A deep detection and target tracking technology, applied in neural learning methods, instruments, biological neural network models, etc., can solve problems such as class imbalance
Active Publication Date: 2020-06-12
JIANGNAN UNIV
View PDF5 Cites 4 Cited by
Summary
Abstract
Description
Claims
Application Information
AI Technical Summary
This helps you quickly interpret patents by identifying the three key elements:
Problems solved by technology
Method used
Benefits of technology
Problems solved by technology
[0007] In order to solve at least one of the above problems, the present invention proposes a long-term object tracking based on deep detection (LT-MDNet) method to solve the problem of occlusion and target out of view during long-term tracking , which can further improve the class imbalance problem during sampling, and enable the model to be effectively updated during online tracking to adapt to changes in the tracking environment, improve long-term target tracking performance, and meet the design requirements of actual engineering systems
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 more
Image
Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
Click on the blue label to locate the original text in one second.
Reading with bidirectional positioning of images and text.
Smart Image
Examples
Experimental program
Comparison scheme
Effect test
Embodiment 1
[0053] Example 1: Long-term target tracking method LT-MDNet based on depth detection
[0054] refer to figure 1 , the specific implementation process of LT-MDNet includes the following steps:
[0055] (1) Offline pre-training, training the weight parameters of three convolutional layers (conv1, conv2, conv3) and two fully connected layers (fc4, fc5) of the MDNet backbone network on the ILSVRC2015-VID target detection tagged dataset;
[0056] (2) Set the last layer of the network (fc6) as a domain-specific layer, which is a binary classification fully connected layer, and output the positive and negative confidence of the sample. The parameters are random at the beginning of each offline training video frame sequence or online tracking video frame sequence initialization.
[0057] (3) Input a new video sequence to be tracked, obtain the first frame of the target (t=1), and manually determine the center position of the target and the length and width of the bounding box (x 1 ...
Embodiment 2
[0074] Embodiment 2: the application of embodiment 1
[0075] 1. Simulation conditions and parameters
[0076] The experiment is implemented based on the PyTorch 1.2.0 programming language and CUDA 10.0 deep learning architecture. The operating system is Windows 10, the processor is AMD R5-2600 3.4GHZ, the GPU is NVIDIA RTX2070, and the memory is 16GB.
[0077] The model is trained offline on the ILSVRC2015-VID target detection label dataset (http: / / bvisionweb1.cs.unc.edu / ilsvrc2015 / ILSVRC2015_VID.tar.gz), and the model parameters are updated every 10 frames; the first frame model updates the training iteration 50 times, the learning rate is 0.0005; the non-first frame update iteration is 15 times, the learning rate is 0.001; the hyperparameters a and c in the loss function are set to 10 and 0.2 respectively, and the shrinkage ratio δ is set to 1.3.
[0078] 2. Simulation content and result analysis
[0079] In order to verify the effectiveness of Example 1 (LT-MDNet), compa...
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 long-time target tracking method based on depth detection, and belongs to the field of pattern recognition and intelligent information processing. According to the method, anMDNet depth detection tracking framework is adopted, and the problem of imbalance of positive and negative samples during sampling is solved by improving a shrinkage loss function on the basis of difficult-to-separate sample mining; designing and maintaining a high-confidence reserved sample pool during online tracking, reserving a first-frame target and high-confidence result sample characteristics, and performing online training by utilizing the reserved sample pool to update model parameters; and finally, calculating confidence coefficients of candidate samples obtained by Gaussian sampling around the target position of the previous frame through the model so as to track the position of the moving target and maintain the robustness of the model through effective updating. According tothe method, excellent tracking precision and success rate are kept in a complex long-term tracking environment, the target position can be accurately positioned when the target is shielded and reappears after the view, and the design requirement of an actual engineering system is met.
Description
technical field [0001] The invention relates to a long-term target tracking method based on depth detection, which belongs to the field of pattern recognition and intelligent information processing. Background technique [0002] With the development of the field of computer vision, object tracking has received more and more attention and applications in the fields of human-computer interaction, video surveillance, automatic driving and robotics. Common early tracking models include particle filter, mean shift, correlation filter and their derivative models. Although the accuracy and speed of these traditional trackers have been significantly improved in recent years, the feature extraction of the target is still based on shallow features such as optical flow, appearance shape, and color, and cannot capture the semantic features of the target. It is difficult to maintain robust tracking in the face of long-term tracking when the target's appearance is deformed, occluded, or ...
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
Application Date:The date an application was filed.
Publication Date:The date a patent or application was officially published.
First Publication Date:The earliest publication date of a patent with the same application number.
Issue Date:Publication date of the patent grant document.
PCT Entry Date:The Entry date of PCT National Phase.
Estimated Expiry Date:The statutory expiry date of a patent right according to the Patent Law, and it is the longest term of protection that the patent right can achieve without the termination of the patent right due to other reasons(Term extension factor has been taken into account ).
Invalid Date:Actual expiry date is based on effective date or publication date of legal transaction data of invalid patent.
Login to view more 
Login to view more