Video Tracking via Apparent Size and Orientation Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking algorithms in video-based systems face challenges such as occlusions, changes in scene illumination, and projective distortions, particularly in transportation scenarios, leading to errors in estimating object size and orientation due to lens distortions like fisheye and telephoto effects, resulting in increased computational complexity.
Innovation Solution
A system and method for video-based tracking that includes a video acquisition unit, object detection unit, object characterization unit, and object localization unit, which utilize a priori knowledge of object size and orientation to adapt tracking parameters, employing morphological operations and computer vision techniques to maintain accurate tracking despite distortions, and adjust the tracking kernel size and orientation based on the object's location and orientation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If appearance-based object tracking is used in video streams, then object tracking can be performed, but errors in estimating object size and orientation occur due to perspective and lens distortion
Solution Approach 1:
The patent transforms the tracking problem from appearance-based to location-based by changing the fundamental parameters used for tracking. Instead of relying on object appearance which is distorted by perspective and lens effects, the system uses the known geometric relationship between camera position and object location to predict expected object size and orientation, thereby resolving the accuracy issues caused by projective distortion
Solution Approach 2:
The patent introduces an intermediate computational step that predicts expected object size and orientation based on location before performing tracking. This intermediary prediction mechanism acts as a mediator that compensates for the distortions, allowing the tracker to use accurate size and orientation information rather than relying directly on distorted appearance measurements
2Loss of information
If object size and orientation are estimated in addition to location, then more comprehensive tracking information is obtained, but computational complexity increases due to higher-dimensional optimization space
Solution Approach 1:
The patent performs preliminary calculation of expected object size and orientation based on the object's location in the scene. By pre-computing these parameters using the known camera geometry and object position, the system avoids the need for complex simultaneous optimization of location, size, and orientation, thereby reducing computational complexity while maintaining complete tracking information
Solution Approach 2:
The patent separates the tracking problem into distinct components: location estimation and size/orientation estimation. By decoupling these tasks and handling them sequentially rather than simultaneously, the system reduces the dimensionality of the optimization problem at each step, making the overall computation more efficient while still obtaining comprehensive tracking information
Data Source
AI summary
A system and method for optimizing video-based tracking of an object of interest are provided. A video of a regularized motion environment that comprise multiple video frames is acquired and an initial instance of an object of interest in one of the frames is then detected. An expected size and orientation of the object of interest as a function of the location of the object is then determined. The location of the object of interest is then determined in a next subsequent frame using the expected size and orientation of the object of interest.