Object Trajectory Classification From Timestamped Tracking Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches for processing sensor data in autonomous vehicles fail to effectively detect physical obstacles in images with occlusion, low resolution, and unusual angles, which is critical for safe navigation.
Innovation Solution
An artificial intelligence model is trained to determine object trajectory by identifying bounding boxes in image sequences, allocating them to tracking identifiers, and generating tracking images with timestamp indications to output trajectory types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional approaches are used for processing sensor data, then the system complexity remains low, but the detection accuracy of physical obstacles in complex images deteriorates
Solution Approach 1:
The patent segments the object detection problem into multiple stages: generating multiple candidate bounding boxes, filtering candidates based on confidence scores, and performing detailed classification on selected candidates. This segmentation allows the system to achieve high detection accuracy while managing computational complexity by processing only relevant regions in detail.
Solution Approach 2:
The patent introduces an additional dimension of processing by generating multiple candidate bounding boxes at different positions and orientations for each detected object. This multi-dimensional approach enables the system to capture objects at various angles and positions, significantly improving detection accuracy for occluded or unusually oriented obstacles.
2Measurement precision
If the system processes all detected objects in detail, then the detection accuracy improves, but the processing time increases
Solution Approach 1:
The patent applies local quality by performing detailed processing only on high-confidence candidate bounding boxes while using simpler filtering for low-confidence candidates. This selective processing strategy maintains high detection accuracy for critical objects while reducing overall processing time by avoiding exhaustive analysis of all detected regions.
Solution Approach 2:
The patent generates multiple candidate bounding boxes (excessive action) for each detected object to ensure comprehensive coverage, but then filters these candidates down to a manageable subset for detailed processing (partial action). This approach ensures no critical objects are missed while maintaining real-time processing capabilities.
3Adaptability or versatility
If multiple candidate bounding boxes are generated for each object, then the detection coverage improves, but the data processing load increases
Solution Approach 1:
The patent performs preliminary filtering of candidate bounding boxes based on confidence scores and spatial relationships before detailed classification. This preliminary action reduces the volume of data requiring intensive processing while maintaining comprehensive detection coverage, as low-confidence or redundant candidates are eliminated early in the pipeline.
Solution Approach 2:
The patent creates multiple candidate bounding boxes (copies) of the same object at different positions and orientations. These copies allow the system to explore multiple possible object interpretations simultaneously, improving detection coverage for ambiguous or partially occluded objects while managing data volume through subsequent filtering of redundant copies.
Data Source
AI summary
Aspects of this technical solution can include identifying, by a processor coupled to non-transitory memory, a plurality of bounding boxes for one or more objects depicted in each image of a sequence of images captured during operation of an autonomous vehicle, allocating, by the processor and based on corresponding positions of the bounding boxes in each image and corresponding time stamps, one or more of the bounding boxes to one or more tracking identifiers each indicating trajectories of corresponding objects, generating, by the processor and based on the time stamps and the bounding boxes allocated to each of the tracking identifiers, one or more tracking images for each of the tracking identifiers, each of the tracking images including visual indications of the time stamps, and training, by the processor and based on the tracking images, an artificial intelligence model to output an indication of a type of trajectory.