Noise-Canceling Learning for False Object Detection in Self-Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional self-driving systems face issues with false detection of road surface features like scratches or repair marks, leading to unnecessary deceleration, as they share and process all captured information, including noise, which can result in useless deceleration when the same place is revisited.
Innovation Solution
A noise-canceling learning device that includes a sensor recognition unit, a detection determination processing unit, and an information storage unit to differentiate between actual and noise objects, associating perceived targets with vehicle positions, allowing for accurate control decisions based on stored information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If all captured sensor information is shared and processed, then information sharing speed is improved, but false detection of noise objects occurs leading to useless deceleration
Solution Approach 1:
The system performs preliminary actions by storing perceived target object information together with position information in advance. When the vehicle returns to the same location, this pre-stored information is retrieved and used to compare with newly detected objects, enabling rapid identification of false detections without re-processing all sensor data from scratch.
Solution Approach 2:
The system implements feedback by comparing newly perceived target objects with previously stored determination target objects at the same position. When a mismatch is detected (indicating a false detection), the system provides feedback to correct the detection error and prevent useless deceleration, thereby improving detection accuracy while maintaining fast information sharing.
2Reliability
If deceleration is performed for every detected suspicious object, then safety is improved, but high-speed traveling becomes difficult due to frequent useless deceleration
Solution Approach 1:
The system performs preliminary comparison of detected objects with pre-stored determination target objects before executing deceleration. This preliminary action filters out false detections (noise objects like road scratches or bridge joint reflections) that would otherwise cause useless deceleration, allowing the vehicle to maintain high-speed travel while still responding to genuine obstacles.
Solution Approach 2:
The system uses feedback from position-based object comparison to determine whether deceleration is necessary. By comparing newly detected objects with previously stored valid objects at the same position, the system provides feedback that confirms whether a detected object is real or noise, enabling safety-conscious deceleration only when truly necessary and maintaining high-speed traveling capability.
3Loss of information
If sensor information is shared through central management device, then information sharing is improved, but only current true information is shared causing repeated false detection at same location
Solution Approach 1:
The system performs preliminary storage of both perceived target objects and determination target objects with position information in advance. This creates a historical record that can be quickly retrieved when the vehicle returns to the same location, eliminating the need to re-share and re-process information through the central management device and preventing repeated false detections and useless deceleration.
Solution Approach 2:
The system implements feedback by comparing newly perceived objects with pre-stored determination target objects at the same position. This feedback mechanism identifies false detections (such as road scratches or bridge joint reflections) and prevents them from being shared as new information, thereby improving information sharing efficiency and eliminating time loss from repeated useless deceleration at the same location.
Data Source
AI summary
Provided is a noise-canceling learning device capable of performing control so as to prevent performing processing such as useless deceleration at the same place in self driving, for example, on a highway, and a vehicle including the noise-canceling learning device. The detection determination processing unit 143 of the noise-canceling learning device 14 causes the perceived object to be stored in (a data table of) the information storage unit 144 as a determination object in association with the perceived object and the own vehicle position information, determines whether the perceived object perceived by the sensor recognition perceiving unit 141 and the determination object stored in (a data table of) the information storage unit 144 match based on the position of the own vehicle 1 estimated by the own vehicle position estimating unit 142, and determines whether the perceived object perceived by the sensor recognition perceiving unit 141 is correct.