Bounding Box Translation for Target Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Prior algorithms for target recognition in imaging devices face errors due to noise from environments, high computational costs, and inefficiencies in tracking targets at low frame rates, especially in crowded or obstructed scenarios.
Innovation Solution
A method that utilizes a reference system to capture both image and depth data streams, synchronizes them, and applies a translation function to predict target positions across frames, allowing for reduced frame rates without increasing tracking errors, and optimizing computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If prior art algorithms use high frame rates to track targets, then target tracking reliability is improved, but computational cost increases
Solution Approach 1:
The patent applies preliminary action by predicting target positions in advance using translation functions before actual detection occurs. The system pre-calculates where targets should appear based on previous frame data and translation models, then only performs full detection when needed. This allows the system to maintain reliable tracking at lower frame rates by having predictions ready beforehand, reducing the computational burden of continuous high-rate detection while preserving tracking accuracy.
2Productivity
If prior art algorithms operate at low frame rates, then computational cost is reduced, but target detection precision deteriorates due to missed overlaps
Solution Approach 1:
The patent uses copying by creating translation functions that replicate target position information from previous frames. Instead of detecting all targets from scratch in each frame, the system copies position data from prior detections and applies translation transformations to predict current positions. This copying mechanism allows low frame rate operation while maintaining detection precision, as the system only needs to verify predictions rather than perform full detection at every frame.
Solution Approach 2:
By pre-calculating target positions using translation functions based on historical data, the system performs preliminary action that compensates for lower frame rates. The translation models predict where targets should be between actual detection frames, maintaining measurement precision even when frames are captured less frequently.
3Reliability
If prior art algorithms require high frame rates to avoid tracking errors, then target tracking reliability is improved, but resource utilization worsens
Solution Approach 1:
The system performs preliminary action by establishing translation functions from historical target position data before actual tracking occurs. These pre-computed translation models allow the system to predict target positions accurately at lower frame rates, maintaining tracking reliability while reducing the energy and computational resources needed for continuous high-rate processing.
4Measurement precision
If prior art algorithms process every frame at high resolution, then target recognition accuracy is improved, but computational cost increases
Solution Approach 1:
The patent applies copying by using translation functions to replicate and transform target position information from previous frames rather than re-detecting all targets in each new frame. This allows the system to maintain recognition accuracy by verifying predictions against actual detections only when necessary, significantly reducing the computational cost of processing every frame at full resolution while preserving measurement precision.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for monitoring (100) targets (10) located in an environment (20) by an imaging device (30), said method comprising the steps of: capturing (110) by the imaging device (30) a first image stream of the environment (20), the first captured image stream comprising a plurality of temporally successive frames (60); sending (120) the first image stream to a data processing unit (40); recognizing (130) one or more targets for each frame (60) of the image stream; associating (140) a bounding box (50) with each recognized target (10); defining (150) a translation function (T) for each bounding box (50) as a function of successive frames (60) of the first captured image stream; associating (180) the defined translation function (T) with each bounding box (50) of a frame (60); translating (190) each bounding box (50) with which a translation function (T) defined in the successive frames (60) is associated by means of the translation function; comparing (200) the translated bounding boxes (70) with the bounding boxes (50) associated with the targets recognized in the frame (60) to which the bounding boxes (70) of the previous frames (60) have been translated; updating (210) the one or more targets (10) recognized in each frame (60) as a function of the comparison.