Deep Learning Target Tracking Model for Real-Time Video Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Region-based Convolutional Neural Networks (R-CNN) based algorithms are inefficient for real-time target tracking in videos due to slow computation speed and struggle to distinguish between targets with similar shapes.
Innovation Solution
An enhanced target tracking model using a hierarchical structure with a feature map generation layer, feature extraction layer via RoI-Align, and a classification layer, along with a new loss function for instance embedding, to improve speed and accuracy in real-time target tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If R-CNN based algorithms are used for target tracking, then target detection capability is improved, but computation speed deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training a target tracking model using pre-inputted training data before actual real-time tracking. The model is pre-trained with target tracking training data to establish initial feature extraction and classification capabilities, enabling fast real-time tracking without performing full R-CNN training during runtime. This separates the computationally intensive training phase from the real-time detection phase.
Solution Approach 2:
The patent extracts and applies only the essential components needed for tracking by using a specialized target tracking model based on CNN architecture. Instead of applying full R-CNN with region proposal generation and multiple classification stages, the patent extracts the core feature extraction and classification functionality adapted for tracking scenarios, removing unnecessary computational overhead.
2Speed
If traditional CNN models are used for tracking, then processing speed is improved, but ability to distinguish similar shapes deteriorates
Solution Approach 1:
The patent applies parameter changes by modifying the CNN model architecture and training parameters specifically for target tracking. The model uses target tracking training data to learn discriminative features that capture subtle differences between similar shapes. The CNN is configured with appropriate layer structures, activation functions, and loss functions optimized for tracking scenarios, enabling both speed and shape distinction capability.
Solution Approach 2:
The patent performs preliminary training of the CNN model with target tracking training data before deployment. This pre-training phase allows the model to learn discriminative features for distinguishing similar shapes while maintaining processing speed. The model is prepared in advance with optimized parameters and feature extraction capabilities specific to tracking tasks.
Data Source
AI summary
The invention disclosed here relates to a method and system for real-time target tracking based on deep learning. The method for real-time target tracking according to an embodiment is performed by a computing device including a processor, and includes pre-training a target tracking model for detecting a tracking target from an image using pre-inputted training data, receiving an image with a plurality of frames, and detecting the tracking target for each of the plurality of frames by applying the target tracking model to the image. According to an embodiment, there is a remarkable reduction in the time required to detect the target from the image, thereby allowing real-time visual tracking, and improvement of the hierarchical structure and introduction of a new loss function make it possible to achieve more precise localization and distinguish different targets of similar shapes.