Siamese Tracking With Structured Dropout for Occlusion Robustness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object trackers face challenges such as appearance variations, occlusion, clutter, and sampling drift, which affect their ability to accurately track objects.
Innovation Solution
The method involves modifying latent space features of a target image using structured dropout techniques, such as channel, segment, and slice dropouts, to mimic occlusion and appearance variations, and aggregating multiple predictions to improve tracking robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional object trackers are used, then the tracking process is simple, but the tracker fails to handle occlusion and appearance variations effectively
Solution Approach 1:
The patent applies preliminary action by performing structured dropout modifications on the target image features before the tracking prediction process. Multiple dropout variants (channel, segment, slice) are pre-generated to simulate occlusion and appearance variations, allowing the tracker to learn robust features in advance rather than reacting to occlusion during tracking.
Solution Approach 2:
The patent segments the feature processing into multiple dropout variants (channel dropout, segment dropout, slice dropout), where each variant processes different portions or aspects of the target image features. This segmentation allows the system to handle different types of occlusion and appearance variations through specialized processing paths.
2Measurement precision
If multiple dropout variants are used to handle occlusion, then tracking accuracy improves, but computational complexity increases
Solution Approach 1:
The patent merges multiple dropout variant predictions through aggregation (e.g., averaging or voting) to produce the final tracking result. This combining approach consolidates the computational effort of multiple dropout variants into a single unified prediction, reducing the overall computational burden while maintaining the accuracy benefits of handling multiple occlusion scenarios.
3Reliability
If structured dropout is applied to target image features, then the tracker becomes more robust to appearance variations, but the processing time increases
Solution Approach 1:
The patent applies partial action by selectively applying different dropout variants based on the specific tracking scenario and occlusion conditions. Rather than always applying all dropout variants, the system can adjust the level of dropout application, using fewer variants when occlusion is minimal and more variants when occlusion is severe, thus optimizing processing time while maintaining robustness.
Data Source
AI summary
A method for object tracking includes receiving a target image of an object of interest. Latent space features of the target image is modified at a forward pass for a neural network by dropping at least one channel of the latent space features, dropping a channel corresponding to a slice of the latent space features, or dropping one or more features of the latent space features. At the forward pass, a location of the object of interest in a search image is predicted based on the modified latent space features. The location of the object of interest is identified by aggregating predicted locations from the forward pass.