Radar Obstacle Detection Using Deep Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional autonomous vehicle perception systems relying on cameras or LIDAR sensors face challenges in detecting obstacles in complex environments with occlusions, inclement weather, and high computational demands, and RADAR systems struggle to distinguish between obstacles and background noise.
Innovation Solution
A deep neural network (DNN) system using RADAR data for object detection, which accumulates and ego-motion-compensates RADAR detections, projects them orthographically, and feeds them into a neural network with a common trunk and multiple heads to predict object presence, location, dimensions, and orientation, enabling differentiation between obstacles and background noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional detection techniques (cameras or LIDAR) are used to detect obstacles, then detection capability in clear conditions is achieved, but reliability deteriorates in scenes with heavy occlusions and inclement weather
Solution Approach 1:
The patent introduces radar sensors as an intermediary detection mechanism that operates effectively in conditions where camera and LIDAR systems fail. The radar system detects obstacles through radio wave reflection, penetrating occlusions and weather conditions that block optical and laser-based sensors, thereby maintaining detection reliability when primary sensors are compromised
Solution Approach 2:
The system employs a composite sensing architecture that integrates radar, camera, and LIDAR sensors. Each sensor type compensates for the weaknesses of others: radar provides all-weather detection capability, cameras provide detailed visual information in clear conditions, and LIDAR provides precise depth mapping. This multi-sensor fusion creates a robust perception system that maintains reliability across diverse environmental conditions
2Reliability
If conventional RADAR detection techniques are used to detect moving objects, then detection of moving reflective objects is achieved, but ability to distinguish stationary objects from background noise deteriorates
Solution Approach 1:
The patent segments the radar detection task into separate processing streams for moving and stationary objects. By dividing the point cloud data and applying different detection algorithms to each segment, the system can optimize parameters for each type: Doppler-based detection for moving objects and spatial-temporal analysis for stationary objects, thereby improving discrimination accuracy for both categories
Solution Approach 2:
The system transitions from traditional two-dimensional radar detection to three-dimensional spatial analysis by integrating depth information from LIDAR and visual data from cameras. This dimensional enhancement allows the system to distinguish stationary objects from background noise through spatial context and geometric consistency across multiple sensor modalities
3Productivity
If conventional RADAR detection techniques are used, then detection of moving objects is achieved, but accuracy in predicting object classification, dimension, and orientation deteriorates
Solution Approach 1:
The patent merges radar detection results with LIDAR point cloud data and camera images to create a unified object representation. By combining the fast moving object detection capability of radar with the precise dimensional and classification data from LIDAR and vision systems, the system achieves both high productivity and accurate object characterization without relying on radar alone
4Reliability
If conventional perception systems (cameras or LIDAR) are used, then obstacle detection is achieved, but computational demands and processing time increase
Solution Approach 1:
The system dynamically adjusts sensor activation and processing intensity based on operational context. In high-risk scenarios with detected obstacles or adverse weather conditions, the system activates all sensors and applies full computational processing. In safe conditions, it reduces processing load by relying on simpler radar-based detection, thereby managing computational complexity while maintaining detection reliability when needed
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 system effectively detects moving and stationary obstacles in complex environments, including inclement weather, with improved accuracy and reduced computational demands compared to conventional methods, enabling safe navigation for autonomous vehicles.
Implementation Method 1
Some conventional techniques use RADAR sensors to detect moving, reflective objects
Data Source
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 some embodiments, ground truth training data for the neural network(s) may be generated from LIDAR data. More specifically, a scene may be observed with RADAR and LIDAR sensors to collect RADAR data and LIDAR data for a particular time slice. The RADAR data may be used for input training data, and the LIDAR data associated with the same or closest time slice as the RADAR data may be annotated with ground truth labels identifying objects to be detected. The LIDAR labels may be propagated to the RADAR data, and LIDAR labels containing less than some threshold number of RADAR detections may be omitted. The (remaining) LIDAR labels may be used to generate ground truth data.