Map-Guided Sensor Data Curation for Rare Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving systems face challenges in efficiently curating large datasets for training machine learning models, particularly for rare objects, due to the time-consuming and labor-intensive process of searching through extensive sensor data for infrequent occurrences.
Innovation Solution
The use of map and geospatial data to filter and identify frames in sensor data that are likely to depict specific objects of interest, combined with deep neural networks for object detection and classification, reduces the dataset size by focusing on high-probability locations and orientations of these objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional methods search through all sensor data frames to find rare objects, then complete data coverage is achieved, but the data curation time and computational resources increase significantly
Solution Approach 1:
The system performs preliminary filtering by querying map data to identify potential locations of rare objects before searching sensor data. This preliminary action narrows down the search space from all frames to only those frames where rare objects are likely to appear, significantly reducing data curation time while maintaining reliability
Solution Approach 2:
Map data serves as an intermediary between the search query and sensor data frames. The system queries map data to obtain potential object locations, which then guides the selection of relevant sensor data frames, acting as a mediator that connects geographic information with actual sensor observations
2Productivity
If map data filtering is applied to identify high-probability frames, then data curation efficiency improves, but system complexity increases due to integration of multiple data sources
Solution Approach 1:
The system segments the data curation process into distinct modules: map data querying, potential location identification, sensor data frame filtering, and object detection. This segmentation allows each component to be independently optimized and managed, reducing overall system complexity while maintaining high productivity
Solution Approach 2:
The system integrates multiple data sources (map data, sensor data) and performs multiple functions (location querying, frame filtering, object detection) within a unified framework. This multi-functional approach improves efficiency by leveraging the complementary strengths of different data sources without requiring separate systems for each function
3Reliability
If large datasets are collected to ensure sufficient training examples for rare objects, then model accuracy potential is maximized, but storage and processing requirements increase
Solution Approach 1:
The system extracts only the relevant frames from the entire sensor data that contain or are likely to contain rare objects. By taking out only the necessary data portions identified through map-based filtering, the system maintains sufficient training examples for accurate model training while dramatically reducing the overall data volume that needs to be stored and processed
Data Source
AI summary
In various examples, map data or geospatial data is used to identify a subset of sensor data having a higher likelihood of including representations of a target object of interest from a larger set of sensor data. Feature vectors corresponding to the subset of sensor data may then be compared to template feature vectors corresponding to the target object in order to confirm the depiction of the target object in the sensor data. The identified sensor data may be used to train one or more machine learning model to compute outputs that correspond to object identification. The trained machine learning models may be used to identify objects in order to aid an autonomous or semi-autonomous machine in a surrounding environment.