Object Trajectory Identification With Filtered Event Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for object identification and trajectory determination perform well in identifying objects but poorly in determining their trajectories, particularly when using RGB images.
Innovation Solution
A method combining image frames with event data analysis, utilizing bounding boxes to filter event data and spiking neural networks for trajectory determination, integrating convolutional neural networks for object identification and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RGB images are used for object identification, then object identification accuracy is improved, but trajectory determination performance deteriorates
Solution Approach 1:
The patent combines RGB image data with event camera data into a unified processing framework. The RGB images provide high-quality object identification while event data provides high-temporal-resolution trajectory information. The system merges these complementary data sources by synchronizing them in time and space, allowing the strengths of each modality to compensate for the other's weaknesses.
Solution Approach 2:
The patent segments the visual information processing into two distinct pathways: one pathway processes RGB images through convolutional neural networks for accurate object identification and classification, while the other pathway processes event data through spiking neural networks for precise trajectory determination. This segmentation allows each pathway to be optimized for its specific function rather than requiring a single system to excel at both tasks.
2Productivity
If event data is filtered based on bounding box position, then processing efficiency is improved, but information completeness may deteriorate
Solution Approach 1:
The patent applies partial filtering by only filtering event data that falls outside the bounding box of identified objects. Event data within the bounding box is retained for trajectory analysis, while data from background regions is discarded. This partial action approach maintains processing efficiency by reducing the total data volume while preserving the essential information needed for trajectory determination.
Solution Approach 2:
The system performs preliminary object identification using RGB images and bounding box detection before filtering the event data. This preliminary action establishes the spatial boundaries (bounding boxes) that guide subsequent event data filtering, ensuring that only relevant event data is processed further. This sequence of operations optimizes processing efficiency by avoiding the need to analyze all event data from the entire field of view.
Data Source
AI summary
A method for identifying objects and determining their trajectories includes analyzing an image frame to identify an object, creating a bounding box around the object, filtering event data based on the position of the bounding box, and analyzing the filtered event data to determine a trajectory of the object. A system for identifying objects and determining trajectories of the objects.