Multi-Target Tracking via Image Synthesis and Hourglass Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current target tracking methods using convolutional neural networks (CNN) are limited in accuracy when the tracking target changes, as they rely solely on a single image frame, leading to decreased performance in subsequent calculations.
Innovation Solution
A multi-target tracking method that involves obtaining two images from the same video file, synthesizing them to create a target image, and then using an hourglass network model for feature extraction, followed by a prediction network to determine updated tracking boxes based on thermodynamic diagrams, width and height values, minor deviation values, and feature vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only a single image frame is used for target tracking, then the device complexity is reduced, but the tracking accuracy deteriorates when the target changes
Solution Approach 1:
The patent merges multiple image frames (current frame and previous frame) into a synthesized image for feature extraction. This combining of temporal information from different frames allows the tracking system to maintain higher accuracy when targets change, while avoiding the need for completely separate processing systems for each frame.
Solution Approach 2:
The patent implements dynamic adaptation by using an hourglass network model that can adjust feature extraction based on the synthesized multi-frame input. The system dynamically adapts to target changes by incorporating temporal information from multiple frames, allowing the tracking parameters to evolve with target variations without increasing overall system complexity.
2Measurement precision
If multiple image frames are synthesized and processed through advanced neural networks, then the tracking accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent segments the complex tracking task into distinct functional modules: frame synthesis module, hourglass network feature extraction module, and prediction module. This segmentation allows each component to be optimized independently, managing computational complexity by breaking down the overall complex processing into manageable stages.
Solution Approach 2:
The patent performs preliminary synthesis of multiple frames into a target image before feature extraction. This preliminary action consolidates temporal information upfront, allowing the subsequent neural network processing to work with a single synthesized image rather than multiple separate frames, thereby reducing the overall computational burden while maintaining accuracy.
3Measurement precision
If a convolutional neural network uses only a single frame as convolution kernel, then the processing speed is maintained, but the accuracy is affected when the tracking target changes
Solution Approach 1:
The patent changes the temporal parameter by incorporating multiple frames into the synthesis process, transforming the input from a single static frame to a dynamically synthesized multi-frame image. This parameter change enables the system to capture target variations over time, improving calculation accuracy without significantly impacting processing efficiency due to the streamlined synthesis approach.
Data Source
AI summary
A multi-target tracking method includes: obtaining a first image and a second image including N tracking targets from the same target video file, and obtaining N first tracking boxes of the first image that precedes a preset number of frames of the second image, the N first tracking boxes configured to frame the N tracking targets in the first image; superposing and synthesizing the first second images to obtain a target image; inputting the target image into an hourglass network model for feature extraction, to output a target feature map; inputting the target feature map into a prediction network to output a thermodynamic diagram, a width and height value set, a minor deviation value set, and a feature vector set; determining N second tracking boxes configured to frame N tracking targets in the second image, based on the thermodynamic diagram, the above three sets, and the N first tracking boxes.