Occlusion Attribute Generation for Autonomous Vehicle Perception
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles (AVs) face challenges in detecting occluded objects due to limitations in sensor perception and the lack of multi-view data integration, which affects their ability to navigate and predict object behaviors accurately.
Innovation Solution
The proposed solution involves a framework that enables AVs to generate occlusion attributes by sharing multi-view data among a fleet of vehicles. This framework allows for the detection of occluded objects and the determination of relational occlusion information and the extent of occlusion, using three-dimensional shapes projected onto a two-dimensional camera space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AVs use single-view sensor data for object detection, then the device complexity is low, but the detection precision for occluded objects deteriorates
Solution Approach 1:
The patent merges sensor data from multiple AVs (first AV and second AV) to detect occluded objects. The object detection module combines detection results from different viewpoints, allowing the system to identify objects that are occluded from any single AV's perspective, thereby improving detection precision without requiring each individual AV to have complex multi-sensor setups.
Solution Approach 2:
The communication module acts as an intermediary that exchanges object detection data between AVs. This mediator enables each AV to access detection results from other AVs, effectively providing multi-view data without requiring each AV to directly process all sensor inputs, thus improving detection capability while maintaining reasonable device complexity.
2Measurement precision
If AVs integrate multi-view data from fleet vehicles, then the detection precision for occluded objects improves, but the loss of time for data communication increases
Solution Approach 1:
The system performs preliminary object detection locally at each AV before exchanging data with the fleet. This preliminary action allows each AV to immediately detect objects in its own field of view without waiting for fleet-wide data aggregation, reducing time loss while still enabling subsequent refinement of occluded object detection through fleet data integration.
Solution Approach 2:
The communication module exchanges only relevant detection data (object positions, types, and detection confidence) rather than complete sensor datasets. This partial action approach shares sufficient information to improve occluded object detection precision while minimizing communication time and data transmission requirements.
3Loss of information
If AVs exchange object detection data with the fleet, then the loss of information about occluded objects is reduced, but the device complexity for data processing increases
Solution Approach 1:
The object detection module extracts only essential object attributes (position, type, detection confidence) from raw sensor data for fleet exchange. This extraction approach minimizes information loss about occluded objects while keeping the exchanged data format simple and easy to process, thereby reducing the data processing complexity burden on each AV.
Solution Approach 2:
The object detection module serves multiple functions: it detects objects for the local AV, prepares data for fleet exchange, and processes incoming fleet data to refine occluded object detection. This multi-functionality reduces the need for separate specialized modules, thereby managing device complexity while comprehensively addressing information loss about occluded objects.
Data Source
AI summary
Autonomous vehicles (AVs) utilize perception and understanding of objects on the road to predict behaviors of the objects, and to plan a trajectory for the vehicle. In some situations, an object may be occluded and undetected by an AV. However, a different AV viewing the same scene may detect the object. With information from multiple views of the same scene, it is possible to determine occlusion attributes for the object, such as relational occlusion information and extent of occlusion. For the AV that is driving on the road, having knowledge of the occluded object and the occlusion attributes can improve the performance of perception, understanding, tracking, prediction, and/or planning algorithms. For the algorithms, occlusion attributes can be generated from the multi-view data and included as part of labeled data for machine learning training. The models in the algorithms can learn to better handle occluded objects.