Vehicle Object Intent Detection with Buffered Multi-Frame Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately determining the intent and behavior of objects in their environment, particularly pedestrians, due to the need for processing multiple image frames to infer actions, which can lead to delayed reaction times and increased computational load.
Innovation Solution
A system utilizing machine learned models that process image data to determine object attributes, including intent, by combining feature vectors from a first model with a vector indicating duplicate frames, allowing for attribute determination with fewer frames and reducing computational load, thereby improving vehicle safety and navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple image frames are processed to determine object intent and behavior, then measurement precision of object attributes is improved, but loss of time increases due to delayed reaction times
Solution Approach 1:
The system performs preliminary processing by pre-computing and storing feature vectors from image frames in a buffer before they are needed for intent determination. When a new frame arrives, the system can immediately compare it with previously buffered feature vectors without waiting for sequential processing, thereby reducing reaction time while maintaining accurate multi-frame analysis
Solution Approach 2:
The system determines object attributes using a subset of available image frames rather than processing all frames sequentially. By selectively using only the necessary number of frames (fewer than traditionally required), the system achieves sufficient measurement precision for intent determination while significantly reducing the time loss associated with processing multiple frames
2Measurement precision
If multiple image frames are processed to determine object intent, then measurement precision is improved, but device complexity increases due to increased computational load
Solution Approach 1:
The system extracts only the essential feature vectors from image frames using a trained machine learning model, storing these compact representations in a buffer. Instead of processing entire image frames or all possible features, the system takes out and processes only the critical attribute information, reducing computational load while maintaining the ability to determine object intent accurately
Solution Approach 2:
The system performs partial processing by analyzing only a subset of image frames rather than all available frames. This selective approach reduces the computational burden significantly while still gathering sufficient information for accurate object attribute determination, as the system can identify when enough frames have been processed to make a reliable intent determination
3Loss of time
If fewer image frames are used to reduce reaction time, then loss of time is reduced, but measurement precision of object attributes deteriorates
Solution Approach 1:
The system performs preliminary computation by pre-extracting and buffering feature vectors from image frames before they are needed for intent determination. This allows the system to immediately begin analyzing frames when needed, reducing reaction time while maintaining the capability to use multiple frames for accurate attribute determination when time permits
Data Source
AI summary
A vehicle computing system may implement techniques to determine attributes (or intent) of an object detected by a vehicle operating in the environment. The techniques may include determining a set of features with respect to a detected object by a first model and determining, by a second model and based on the set of features, one or more attributes of the object. The first model and the second model may be configured to process at least one image frame to determine the one or more attributes of the object. A model may receive sensor data as an input, and output features and/or an attribute for the detected object. Based on the attribute(s) of the object, a vehicle computing system may control operation of the vehicle.