3D Motion Detection Using Voxel Occupancy Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion detection systems for autonomous vehicles struggle to accurately detect small, instantaneous motions in traffic scenes due to noise in object localization data, which can lead to collisions or accidents.
Innovation Solution
A method and system for motion detection that involves performing two predictions based on a sequence of input point clouds from LiDAR sensors, including a first prediction to predict voxel occupancy and a second prediction using predicted voxel occupancy between input point clouds, followed by motion detection and responsive actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object localization data from object detection and tracking is used for motion detection, then large motions are relatively easy to identify, but fine-grained small motion becomes difficult to detect due to noise in the data
Solution Approach 1:
The patent segments the motion detection task into two distinct prediction processes: a first prediction for voxel occupancy and a second prediction for voxel occupancy using predicted voxel occupancy between input point clouds. This segmentation allows each prediction to focus on specific aspects of motion, improving overall detection precision for fine-grained motions while maintaining reliability through the divided approach.
Solution Approach 2:
The patent introduces predicted voxel occupancy as an intermediary element between the input point clouds and the final motion detection. This intermediary representation filters out noise in object localization data while preserving fine-grained motion information, enabling reliable detection of small motions that would otherwise be lost in noisy data.
2Productivity
If traditional motion detection methods are used, then the system complexity remains low, but the ability to detect small instantaneous motions in real-time is insufficient
Solution Approach 1:
The patent performs preliminary predictions of voxel occupancy before final motion detection. The first prediction establishes baseline voxel occupancy, and the second prediction refines this using predicted voxel occupancy between input point clouds. These preliminary actions prepare the data structure in advance, enabling real-time detection of small motions without excessive computational complexity during the actual detection phase.
Solution Approach 2:
The patent transforms the motion detection problem from direct point cloud analysis to voxel occupancy space. By converting point cloud data into voxel representations and performing predictions in this transformed dimensional space, the system achieves real-time detection capability while managing complexity through the structured voxel framework.
Data Source
AI summary
Methods and systems for motion detection include performing a first prediction to predict voxel occupancy based on a sequence of input point clouds including a current point cloud and a set of previous point clouds. A second prediction is performed to predict voxel occupancy for the sequence of input point clouds using predicted voxel occupancy between the input point clouds. Motion detection is performed based on the completed voxel occupancy. An action is performed responsive to a detected motion.