Radar Obstacle Detection Using DNN Instance Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional perception systems for autonomous vehicles rely heavily on cameras and LIDAR, which are unreliable in occluded scenes and inclement weather, and RADAR systems struggle to distinguish between obstacles and background noise, especially in cluttered urban environments.
Innovation Solution
A deep neural network (DNN) using RADAR data is trained to detect moving and stationary obstacles by accumulating, ego-motion-compensating, and orthographically projecting RADAR detections, with a common trunk and separate heads for class confidence and instance regression, and uses LIDAR data for ground truth training to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional RADAR detection techniques are used to detect moving objects, then detection capability for moving reflective objects is improved, but the ability to distinguish stationary objects from background noise deteriorates
Solution Approach 1:
The detection task is segmented into two separate specialized detectors: one for moving objects and one for stationary objects. Each detector is optimized for its specific detection target, allowing the system to maintain high reliability for moving object detection while simultaneously achieving high precision in distinguishing stationary objects from background noise, eliminating the trade-off between these two capabilities.
Solution Approach 2:
The system dynamically adapts detection strategies based on object motion state. By categorizing objects as either moving or stationary, the system applies appropriate detection algorithms for each category, enabling reliable detection of moving reflective objects while accurately distinguishing stationary objects from background clutter through motion-based differentiation.
2Reliability
If conventional perception systems use cameras or LIDAR sensors, then detection capability in clear conditions is improved, but reliability in occluded scenes and inclement weather deteriorates
Solution Approach 1:
The patent introduces RADAR technology as an intermediary sensing modality that bridges the gap between optical sensors (cameras/LIDAR) and the challenging environmental conditions. RADAR serves as a mediator that can penetrate occlusions and weather obstacles, providing reliable detection capability where conventional optical-based systems fail, thus resolving the contradiction between detection performance and environmental sensitivity.
3Device complexity
If conventional detection techniques are used, then system complexity is reduced, but measurement precision and run-time efficiency deteriorate due to heavy post-processing requirements
Solution Approach 1:
The system performs preliminary classification of objects as moving or stationary during the detection phase itself, before any post-processing occurs. This preliminary action embedded in the detection algorithm eliminates the need for subsequent heavy post-processing to extract 3D information, thereby maintaining measurement precision while reducing overall system complexity and improving run-time efficiency.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The DNN effectively distinguishes between obstacles and background noise, enabling reliable obstacle detection in complex urban environments and inclement weather conditions, improving the safety and efficiency of autonomous vehicle navigation.
Implementation Method 1
RADAR sensors 1360 may be used to generate RADAR data 102
Implementation Method 2
ground truth data 109 may be generated from LIDAR data 103
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In various examples, a deep neural network(s) (e.g., a convolutional neural network) may be trained to detect moving and stationary obstacles from RADAR data of a three dimensional (3D) space, in both highway and urban scenarios. RADAR detections may be accumulated, ego-motion-compensated, orthographically projected, and fed into a neural network(s). The neural network(s) may include a common trunk with a feature extractor and several heads that predict different outputs such as a class confidence head that predicts a confidence map and an instance regression head that predicts object instance data for detected objects. The outputs may be decoded, filtered, and/or clustered to form bounding shapes identifying the location, size, and/or orientation of detected object instances. The detected object instances may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.