Long-Range LiDAR Speed Estimation Using Segmented Range CNNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing long-range speed estimation systems in ADAS and AD environments face inaccuracies due to dependence on inertial and GPS sensors, and camera-based object annotation and tracking, which is challenging at long distances, impacting vehicle safety.
Innovation Solution
The system divides LIDAR frames into short, middle, and long ranges, using convolutional neural networks to estimate self and relative speeds, with a hand-over mechanism to combine results, and optionally includes an optical flow estimation network for accurate speed determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera-based object annotation and tracking is used to determine relative speed, then speed estimation can be performed, but detection accuracy deteriorates at long distances
Solution Approach 1:
The system segments the LIDAR point cloud data into multiple depth ranges (e.g., near, mid, far ranges) and processes each range separately with specialized networks. This segmentation allows the system to optimize detection strategies for different distance zones, improving overall long-range detection accuracy by treating distant objects with dedicated processing rather than a unified approach.
Solution Approach 2:
The system introduces LIDAR data as an intermediary source to complement camera-based tracking. By fusing LIDAR point cloud information with camera imagery, the system obtains more reliable depth and motion data for distant objects, overcoming the camera's limitation in tracking objects at long distances where visual features become indistinct.
2Measurement precision
If inertial and GPS sensors are used to determine self speed, then speed measurement can be performed, but measurement accuracy deteriorates due to sensor dependence
Solution Approach 1:
The system merges multiple speed estimation approaches by combining self-speed estimation from LIDAR point cloud analysis with relative speed estimation from object tracking. This fusion of multiple independent measurement sources creates a more reliable absolute speed calculation that compensates for the weaknesses of individual sensors and reduces overall system dependence on any single sensor type.
Solution Approach 2:
The system enables the LIDAR sensor to serve multiple functions: it not only provides primary object detection and range measurement but also performs self-speed estimation by analyzing the motion of detected objects and the ego-vehicle's own point cloud changes over time. This self-service capability reduces reliance on external inertial and GPS sensors for speed measurement.
3Device complexity
If LIDAR frames are processed as a whole, then processing is simpler, but computation certainty deteriorates at long distances
Solution Approach 1:
The system divides LIDAR frames into multiple depth-range segments and processes each segment independently with specialized convolutional neural networks. This segmentation increases processing complexity but significantly improves computation certainty for long-distance objects by dedicating specific network configurations to different range zones, allowing each segment to be optimized for its specific detection challenges.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A LIDAR-based method of determining an absolute speed of an object at a relatively longer distance from an ego vehicle, including: estimating a self speed of the ego vehicle using a first frame t-1 and a second frame t obtained from a LIDAR sensor by estimating an intervening rotation θ about a z axis and translation in orthogonal x and y directions using a deep learning algorithm over a relatively closer distance range; dividing each of the first frame t-1 and the second frame t into multiple adjacent input ranges and estimating a relative speed of the object at the relatively longer distance by subsequently processing each frame using a network, with each input range processed using a corresponding convolutional neural network; and combining the estimation of the estimating the self speed with the estimation of the estimating the relative speed to obtain an estimation of the absolute speed.