Autonomous Vehicle Object Detection via Joint Classifier
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in processing real-time stereoscopic camera images to identify and classify obstacles efficiently, and adding additional sensors increases processing power requirements and generates redundant data.
Innovation Solution
A joint classifier system that processes sensor data from stereo cameras and optical flow sensors to create disparity maps and optical flow images, comparing them to 3D environment data to identify unknown objects and classify hazards such as pedestrians, vehicles, and bicyclists, reducing unnecessary data processing by focusing on interesting aspects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If stereoscopic camera images are processed in real-time to identify and classify obstacles, then hazard identification capability is improved, but processing power requirements increase
Solution Approach 1:
The patent extracts and processes only the most relevant features from sensor data rather than analyzing complete images. The joint classifier system identifies and processes only interesting aspects of the scene, removing unnecessary data processing while maintaining hazard identification capability.
Solution Approach 2:
The system performs partial processing by focusing computational resources only on regions of interest identified by the joint classifier. Instead of processing all sensor data equally, it applies selective processing to areas that contain potential hazards, reducing overall processing power requirements while maintaining detection reliability.
2Measurement precision
If additional sensors are added to improve detection capability, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent merges data from multiple sensor types (stereo cameras and optical flow sensors) into a unified processing framework. The joint classifier system combines information from both sensor modalities to achieve improved detection precision while managing system complexity through integrated processing rather than separate analysis channels.
Solution Approach 2:
The joint classifier system serves multiple functions simultaneously: it processes both stereo vision data and optical flow data, performs depth estimation, motion detection, and object classification. This multi-functional approach allows the system to achieve comprehensive detection capability without requiring separate dedicated processing systems for each sensor type.
3Measurement precision
If complete sensor data is processed to ensure accurate object classification, then classification accuracy is improved, but processing time increases
Solution Approach 1:
The joint classifier performs preliminary analysis to identify interesting aspects and regions of interest before detailed object classification is performed. This preliminary filtering step prepares the data in advance, allowing subsequent classification to focus only on relevant objects and thereby reducing overall processing time while maintaining accuracy.
Solution Approach 2:
The processing system segments the sensor data into different components and processes them through specialized pathways. The joint classifier divides the scene into regions of interest and non-interest areas, processing only the relevant segments in detail while using simplified processing for other areas, thus reducing total processing time while preserving classification accuracy for critical objects.
Data Source
AI summary
An object detection system for an autonomous vehicle processes sensor data, including one or more images, obtained for a road segment on which the autonomous vehicle is being driven. The object detection system compares the images to three-dimensional (3D) environment data for the road segment to determine pixels in the images that correspond to objects not previously identified in the 3D environment data. The object detection system then analyzes the pixels to classify the objects not previously identified in the 3D environment data.