Autonomous Vehicle Object Classification Using Gated Camera Optical Flow
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle systems struggle to accurately determine whether detected objects can be driven over, leading to potentially unnecessary evasive maneuvers or braking, as they rely on object position, size, and mass, which is not always reliable.
Innovation Solution
A method using a gated camera to record images of the roadway and determine optical flow, considering self-motion information, to classify objects as able to be driven over or not, thereby deciding on necessary maneuvers like braking or evading.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object classification is based on position, size, and mass, then the system can detect objects, but the classification accuracy is insufficient leading to unnecessary maneuvers
Solution Approach 1:
The patent introduces a new classification dimension beyond position, size, and mass by analyzing object movement patterns and road contact status. The system evaluates whether objects are stationary or moving, and whether they contact the roadway, adding temporal and spatial behavior dimensions to the classification process. This multi-dimensional approach improves classification accuracy and reduces false positives in maneuver triggering.
2Reliability
If the system performs evasive maneuvers or braking for all detected objects, then collision avoidance is prioritized, but unnecessary maneuvers increase and reduce operational efficiency
Solution Approach 1:
The patent applies different response strategies based on local characteristics of each detected object. Instead of a uniform response to all objects, the system classifies objects into different categories (stationary vs. moving, road-contacting vs. non-road-contacting) and applies appropriate maneuver decisions for each category. This localized differentiation maintains high collision avoidance reliability while reducing unnecessary maneuvers for objects that pose minimal risk.
3Difficulty of detecting and measuring
If traditional sensor-based object detection is used, then objects can be detected, but the ability to determine drivability over objects is insufficient
Solution Approach 1:
The patent performs preliminary classification of objects based on movement patterns and road contact status before making final drivability assessments. By first determining whether objects are stationary or moving and whether they contact the roadway, the system prepares classification information in advance that guides subsequent drivability evaluation. This preliminary action improves the precision of drivability assessment by providing additional contextual information about object behavior.
Data Source
AI summary
A gated camera records images of a roadway in a longitudinal direction in front of a vehicle. An object is recognized in the images and it is determined using an optical flow, considering distinct movement information of the vehicle, whether a braking and/or an evading maneuver is required to avoid a collision with the object. Using the optical flow it is determined whether the object is in contact with the roadway, has a distinct movement, is moved out of a driving corridor of the vehicle or into the driving corridor. When the distinct movement of the object is erratic or the object is moved out of the driving corridor, the object is classified as able to be driven over or otherwise as not able to be driven over. A braking and/or evading maneuver is carried out only with an object that is classified as not able to be driven over.