Ultrasonic Object Detection Using 3D Point Cloud Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ultrasonic sensors face challenges in object detection and localization due to the lack of directionality information, low sampling rate, and signal noise, making them less effective compared to other sensors like cameras, LiDARs, and Radars for autonomous vehicles, despite their suitability for near-range detection in harsh weather conditions.
Innovation Solution
A grid-based 2D bird's eye view transformation approach is adapted for ultrasonic data to generate a 3D point cloud representation, which is then projected onto a 2D plane encoding echo-intensity, amplitude, and azimuth-angle information, enabling the training of machine-learning models for object detection and localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If ultrasonic sensors are used for object detection, then cost is reduced and near-range detection capability is improved, but directionality information is lost and measurement precision deteriorates
Solution Approach 1:
The patent transforms 1D ultrasonic echo data into 2D bird's-eye view representations by introducing spatial dimensionality. Multiple ultrasonic sensors arranged in arrays capture echoes from different positions, and these 1D distance measurements are projected onto a 2D plane to reconstruct spatial information about objects, effectively compensating for the lack of inherent directionality in ultrasonic sensing.
Solution Approach 2:
The patent divides the sensing task into multiple segments by using multiple ultrasonic sensors positioned at different locations. Each sensor provides a segment of the overall spatial picture, and by combining measurements from multiple sensors, the system reconstructs complete object information including position and orientation that would be unavailable to a single sensor.
2Device complexity
If traditional ultrasonic sensing methods are used, then device complexity is reduced, but object detection accuracy deteriorates due to signal noise and low sampling rate
Solution Approach 1:
The patent merges data from multiple ultrasonic sensors with multiple reflections into a unified 2D bird's-eye view representation. By combining echoes from multiple sensors and multiple reflections off the same object, the system accumulates sufficient statistical information to achieve reliable object detection and localization despite individual measurements being noisy.
Solution Approach 2:
The patent performs preliminary processing of ultrasonic echo data by transforming it into 2D bird's-eye view representations before object detection. This preprocessing step organizes the noisy 1D echo measurements into structured 2D spatial information, making subsequent object detection more robust to noise and enabling the use of standard computer vision algorithms.
3Adaptability or versatility
If camera-based object detection is used, then adaptability to varying light conditions is improved, but performance deteriorates when scenes are blocked by obstacles such as fog or snow
Solution Approach 1:
The patent replaces optical sensing (cameras) with acoustic sensing (ultrasonic sensors) for object detection. Ultrasonic waves are not blocked by fog, snow, or dust in the same way light is, allowing the system to maintain reliable detection capability in adverse weather conditions where camera-based systems fail.
4Length of stationary object
If LiDAR is used for object detection, then long-range detection capability is improved, but cost increases and performance deteriorates in adverse weather conditions
Solution Approach 1:
The patent uses inexpensive ultrasonic sensors instead of expensive LiDAR systems. While individual ultrasonic sensors have shorter effective ranges than LiDAR, the system compensates by using multiple sensors in arrays and leveraging multiple reflections to extend effective detection range while maintaining low cost.
5Reliability
If Radar is used for object detection, then performance in varying weather conditions is improved, but resolution deteriorates and near-range performance is poor
Solution Approach 1:
The patent optimizes ultrasonic sensing for near-range detection by arranging sensors in dense arrays and using multiple reflections to enhance local spatial resolution. The system is specifically tuned for short-to-medium range detection where ultrasonic waves maintain sufficient energy, achieving high resolution in the near field where LiDAR and Radar are less cost-effective.
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
This approach enables accurate object detection in low-speed scenarios, improving the accuracy and safety of perception tasks for autonomous vehicles by leveraging the resilience and low-cost of ultrasonic sensors, even in adverse weather conditions.
Implementation Method 1
object detection and localization using a plurality of ultrasonic sensors embedded on a moving vehicle. During each of a plurality of cycles, the method may include simultaneously emitting, according to a fixed pattern, an ultrasonic signal from one or more of the plurality of ultrasonic sensors, wherein each emitted ultrasonic signal is received as an echo at one of the plurality of ultrasonic sensors
Data Source
AI summary
A novel ultrasonic-sensor based object detection framework, useful for autonomous vehicles, is disclosed. Disclosed embodiments input tabular ultrasonic data into a 3D point cloud. Multi-channel, bird's eye view data structures are obtained from the 3D point cloud. High quality, ultrasonic images are generated using a temporal-aggregation method and used as input to a machine-learning model.