Autonomous Vehicle Environmental Change Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face challenges in efficiently detecting changes in the driving environment, such as potholes and moving objects, due to high processing demands and inaccuracies in machine learning algorithms, and require extensive mapping and infrastructure, which is impractical and costly.
Innovation Solution
The system employs road side units (RSUs) with radio frequency capabilities to provide a background template image, allowing for efficient motion detection and obstacle identification using image subtraction and minimal processing power, with optional support from DNN/CNN and Lidar/Radar for precise object recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms (CNN/DNN) are used for object detection, then object recognition capability is improved, but processing power requirements and computational cost increase significantly
Solution Approach 1:
The system segments the detection task into two parts: a lightweight initial filter that processes all frames to detect potential changes, and a more sophisticated ML-based analyzer that only processes detected anomalies. This segmentation allows accurate object recognition only when necessary, reducing overall computational load while maintaining detection accuracy.
Solution Approach 2:
Instead of applying full ML-based object recognition to every video frame, the system applies partial action by first using a simple change detection algorithm to identify only the frames containing potential objects or changes. ML algorithms are then applied only to these selected frames, significantly reducing processing requirements while maintaining recognition accuracy.
2Measurement precision
If detailed high-resolution maps are created for autonomous navigation, then navigation accuracy is improved, but mapping cost and infrastructure requirements increase
Solution Approach 1:
The system uses the autonomous vehicle's own onboard sensors (cameras, LIDAR, GPS) to continuously build and update its environmental map in real-time, eliminating the need for pre-existing high-resolution maps. The vehicle serves its own mapping needs by processing sensor data locally and creating navigation-relevant features on-the-fly.
Solution Approach 2:
The system performs preliminary actions by pre-processing sensor data to extract only the essential navigation features (road boundaries, obstacles, landmarks) needed for autonomous driving, rather than creating complete high-resolution maps of the entire environment. This selective feature extraction reduces mapping complexity while maintaining navigation accuracy.
3Productivity
If crowd-sourced map updates are used, then map update frequency is improved, but data reliability decreases due to intentional inaccuracies
Solution Approach 1:
The system relies on its own onboard sensors to independently verify and update map data, rather than depending on crowd-sourced information from other vehicles or sources. This self-service approach ensures data reliability by using direct sensor observations from the autonomous vehicle itself, while still maintaining frequent updates through continuous sensor processing.
4Measurement precision
If LIDAR is used for environmental sensing, then detection precision is improved, but system cost increases significantly
Solution Approach 1:
The system merges multiple lower-cost sensor types (cameras, LIDAR when available, GPS, IMU) into a unified sensing system that achieves LIDAR-level detection precision through sensor fusion. By combining the strengths of different sensors and using ML-based data fusion algorithms, the system attains high detection accuracy without requiring expensive LIDAR hardware in all configurations.
Solution Approach 2:
The system uses computer vision algorithms to create a virtual copy of the 3D environmental model that would otherwise require physical LIDAR to generate. By processing 2D camera images through sophisticated ML algorithms, the system reconstructs 3D spatial relationships and object characteristics, providing LIDAR-equivalent detection capability through software rather than expensive hardware.
Data Source
AI summary
In some embodiments, the disclosed subject matter involves a system and method for dynamic object identification and environmental changes for use with autonomous vehicles. For efficient detection of changes for autonomous, or partially autonomous vehicles, embodiments may use a technique based on background removal and image subtraction which use motion detection rather than full object identification for all objects in an image. Road side units proximate to a road segment or virtual road side units in the cloud, other vehicles or mobile device (e.g., drones) are used to retrieve and store background images for a road segment, to be used by the autonomous vehicle. Other embodiments are described and claimed.