Unmanned Vehicle Object Detection Box Validation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for generating object detection boxes in unmanned vehicles using point-cloud data are labor-intensive and lack accuracy, particularly in automatic detection algorithms.

Innovation Solution

A method and apparatus that obtain point-cloud data frames from a radar device, determine the closest approach time to a target object, and validate detection box information using automatic algorithms to ensure accurate and stable detection box generation by smoothing backward-checked results.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual annotation method is used to obtain detection box information, then accuracy of detection box generation is improved, but labor cost and time consumption increase significantly

Engineering Contradiction:

Improveaccuracy of detection box generationVSAvoidtime consumption

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by performing automatic detection on multiple point-cloud data frames in advance, then using the results from the closest approach frame to validate and correct detection results in previous frames. This preliminary processing enables subsequent frames to benefit from improved accuracy without requiring manual annotation for each frame, thus reducing time consumption while maintaining accuracy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent uses detection box information from the first point-cloud data frame (where the vehicle is closest to the target) as a reference copy to validate and correct detection results in the second point-cloud data frame. By copying and comparing detection results across different frames, the system achieves high accuracy without manual annotation for each frame, resolving the contradiction between accuracy and time consumption.

Inventive Principle:

Principle #26Copying

2Productivity

If automatic detection algorithm is used to obtain detection box information, then labor cost is reduced, but accuracy and reliability of detection results deteriorate

Engineering Contradiction:

Improvelabor efficiencyVSAvoidaccuracy of detection box generation

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent implements feedback by using detection box information from the first point-cloud data frame (obtained via automatic detection) to validate and correct detection results in the second point-cloud data frame. The system feeds back the more reliable detection results from the closest approach frame to improve the accuracy of previous frame detections, thereby maintaining high productivity while improving accuracy through iterative validation.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent merges detection results from multiple point-cloud data frames by combining automatic detection algorithms with validation against reference frames. By merging information from the first detection box (high reliability) with the second detection box (needs validation), the system achieves both high productivity and improved accuracy, resolving the contradiction between automated processing and result reliability.

Inventive Principle:

Principle #5Merging (Combining)

3Measurement precision

If detection box information from closest approach frame is used to validate previous frames, then accuracy of historical detection results is improved, but processing complexity increases

Engineering Contradiction:

Improveaccuracy of historical detection resultsVSAvoidprocessing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by first identifying the point-cloud data frame where the vehicle is closest to the target object, then using this frame's detection results as a reference for validating previous frames. This preliminary identification simplifies the overall process by establishing a single reference point, reducing processing complexity while improving accuracy of historical detections.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent uses an inverted approach by not validating the closest approach frame against previous frames, but rather using the closest approach frame to validate previous frames. This inversion simplifies processing because the closest approach frame provides the most reliable detection data, eliminating the need for complex iterative validation across all frames and reducing overall processing complexity.

Inventive Principle:

Principle #13The other way round (Inversion)

Data Source

PatentUS11415672B2Method and apparatus for generating object detection box, device, storage medium, and vehicle

Publication Date: 2022.08.16 APOLLO INTELLIGENT DRIVING (BEIJING) TECHNOLOGY CO LTD

AI summary

The present disclosure provides a method for generating an object detection box, comprises: obtaining a set of point-cloud data frames collected by a radar device within a set period; obtaining, from the set, a first point-cloud data frame corresponding to a first time when an unmanned vehicle is closest to a target object, and obtaining first detection box information corresponding to the target object in the first point-cloud data frame that is obtained through an automatic detection algorithm; determining whether the first detection box information is valid detection box information with respect to a second cloud-point data frame in the set that corresponds to a second time prior to the first time; and determining whether the first detection box information is to be used as final detection box information of the target object in the second point-cloud data frame according to the result of the determination.