Vehicle Maneuver Classification With Spatiotemporal Attention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for classifying vehicle maneuvers fail to utilize GPS and IMU data, leading to inaccurate classifications, excessive resource consumption, and potential dangerous maneuvers due to incorrect classifications.
Innovation Solution
A video system that classifies vehicle maneuvers using spatiotemporal attention selectors, processing video, GPS, and IMU data to generate feature vectors and identify safe or unsafe maneuvers, conserving resources by accurately linking objects across frames and selecting relevant features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current techniques for classifying vehicle maneuvers are used, then the system operates with simpler processing, but the classification accuracy deteriorates due to failure to utilize GPS and IMU data
Solution Approach 1:
The patent combines multiple data sources (video frames, GPS data, and IMU data) into a unified classification system. The spatiotemporal attention selector integrates features from video processing, GPS location tracking, and IMU motion sensing to jointly determine maneuver classification, thereby improving accuracy while managing complexity through integrated processing.
Solution Approach 2:
The system segments the classification task into distinct processing streams: video frame processing, GPS data processing, and IMU data processing. Each stream extracts relevant features independently, which are then combined by the spatiotemporal attention selector. This segmentation allows complex multi-source data to be processed in manageable parts while maintaining overall system accuracy.
2Reliability
If current techniques are used without GPS and IMU data integration, then the system consumes fewer resources, but the maneuver classification becomes inaccurate leading to dangerous situations
Solution Approach 1:
The spatiotemporal attention selector applies attention mechanisms selectively to different data sources based on their relevance to the current classification task. Rather than processing all features from GPS and IMU data uniformly, the system dynamically weights and selects only the most pertinent features, reducing unnecessary computing resource consumption while maintaining classification reliability.
Solution Approach 2:
The system performs preliminary processing of GPS and IMU data to extract relevant features before combining them with video features. This preliminary action prepares the data in advance, reducing the computational burden during the final classification stage and improving overall system efficiency while maintaining reliability.
3Measurement precision
If video data alone is processed without linking objects across frames, then the processing is simpler, but the classification accuracy deteriorates due to loss of temporal context
Solution Approach 1:
The system dynamically tracks objects across video frames, adapting to changing scenes and maneuvers. The spatiotemporal attention selector dynamically adjusts which object features to emphasize based on temporal context from previous frames, enabling accurate classification of evolving maneuvers while managing tracking complexity through adaptive processing.
Data Source
AI summary
A device may receive video data and corresponding GPS data and IMU data associated with a vehicle, and may process the video data, with an object detector model, to identify objects and to generate a first feature vector. The device may process the GPS data and the IMU data, with a first CNN model, to generate a second feature vector, and may process the objects and the video data, with a tracking model, to identify positions and classes of the objects and to generate a third feature vector. The device may utilize a second CNN model to generate a matrix of object features based on the first, second, and third feature vectors, and may utilize a spatiotemporal attention selector model or a max pooled model with the matrix of object features to identify a classification of a maneuver of the vehicle. The device may perform actions based on the classification.