Object Tracking via Multi-Layer CNN Feature Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking methods face challenges in handling significant appearance changes, pose variations, and background clutters due to reliance on low-level hand-crafted features that lack semantic information and discriminative power.
Innovation Solution
The method employs a convolutional neural network (CNN) to select and utilize feature maps from higher and lower layers, with a general network (GNet) capturing semantic features and a specific network (SNet) providing discriminative information to separate targets from distracters, using pre-trained networks for robust tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If low-level hand-crafted features are used for object tracking, then the tracking system is simpler to implement, but the system lacks robustness to significant appearance changes and has limited discriminative power
Solution Approach 1:
The CNN is pre-trained on large-scale image classification datasets (e.g., ImageNet) before being applied to object tracking. This preliminary training enables the network to learn robust feature representations that generalize well to tracking tasks, resolving the contradiction by preparing the system in advance to handle appearance variations without increasing operational complexity
Solution Approach 2:
The patent transforms the feature extraction approach by changing from fixed hand-crafted features to adaptive learned features through CNN. The network learns optimal feature parameters automatically from data, improving robustness to appearance changes while the modular CNN architecture keeps implementation manageable
2Loss of information
If low-level hand-crafted features are used for object tracking, then the computational requirements are lower, but the features lack semantic information and have limited discriminative power
Solution Approach 1:
The patent segments the feature extraction process into multiple CNN layers, where lower layers capture basic visual patterns and higher layers progressively learn more abstract semantic representations. This hierarchical segmentation allows the system to retain rich semantic information while managing computational load through progressive feature learning
Solution Approach 2:
The patent transitions from traditional 2D image processing to multi-dimensional feature space by utilizing deep CNN layers. This dimensional transformation enables the extraction of high-level semantic features that capture complex object properties, improving discriminative power despite increased computational requirements
3Loss of information
If features from higher CNN layers are used, then more semantic information is captured, but less discriminative information is available to separate objects from distracters with similar appearance
Solution Approach 1:
The patent merges feature maps from multiple CNN layers by concatenating or combining them into a unified feature representation. This merging strategy integrates both high-level semantic information from deeper layers and low-level discriminative details from shallower layers, achieving a balance between semantic richness and separation accuracy
Solution Approach 2:
The patent creates a composite feature representation by combining features from different CNN layers, analogous to composite materials that combine properties of constituent materials. This composite feature vector leverages the strengths of both high-level and low-level features to achieve robust object separation while maintaining semantic information
4Measurement precision
If features from lower CNN layers are used, then more discriminative information is available to separate objects from distracters, but less semantic information is captured
Solution Approach 1:
The patent merges feature maps from multiple CNN layers by concatenating or combining them into a unified feature representation. This merging strategy integrates both high-level semantic information from deeper layers and low-level discriminative details from shallower layers, achieving a balance between semantic richness and separation accuracy
Solution Approach 2:
The patent segments the feature extraction process into multiple CNN layers, where lower layers capture basic visual patterns and higher layers progressively learn more abstract semantic representations. This hierarchical segmentation allows the system to retain rich semantic information while managing computational load through progressive feature learning
Data Source
AI summary
Disclosed are methods for object tracking. In an example, the method comprises: determining a region of interest (ROI) in a first frame of a video sequences; feeding the determined ROI forward through a first CNN (convolutional network) to obtain a plurality of first feature maps in a higher layer of the CNN and a plurality of second feature maps in a lower layer of the first CNN; selecting a plurality of feature maps from the first and second feature maps, respectively; predicting, based on the selected first and second feature maps, two target heat maps indicating a target location for said objects in the current frame, respectively; and estimating, based on the two predicated target heat maps, a final target location for the object in the current frame.