Object Identification Device Using Weighted Sensor Confidence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object identification devices inaccurately identify pedestrians when they are positioned beyond the detectable range of laser radars due to fluctuation in reflection levels, leading to incorrect classification.
Innovation Solution
An object identification device employing multiple sensors that output confidence levels for object categorization, with a processing unit calculating an average confidence to accurately identify objects by integrating radar and camera data, using weighted averages to enhance identification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a laser radar is used to detect objects beyond the pedestrian detectable region, then the detection range is extended, but the identification accuracy deteriorates due to fluctuation in reflection levels
Solution Approach 1:
The patent combines detection results from multiple sensors (laser radar and camera) to perform object identification. The laser radar detects objects in extended ranges while the camera provides verification for ambiguous cases, merging their capabilities to achieve both extended detection range and maintained identification accuracy
Solution Approach 2:
The patent introduces a camera as an intermediary verification device. When the laser radar detects an object beyond its reliable identification range, the camera acts as an intermediary to perform pattern matching and confirm whether the object is a pedestrian, thus resolving the accuracy deterioration problem
2Device complexity
If a single sensor is used for object identification, then the device complexity is reduced, but the identification reliability deteriorates due to inability to handle ambiguous detection cases
Solution Approach 1:
The patent segments the identification process into two stages: first, the laser radar performs initial detection and identifies obvious objects; second, the camera performs verification on ambiguous cases. This segmentation allows each sensor to focus on its strengths while maintaining system reliability
Solution Approach 2:
The patent applies partial action by having the camera process only those objects that the laser radar identifies as ambiguous or beyond its reliable range, rather than having the camera process all detected objects. This reduces unnecessary processing while maintaining reliability
Data Source
AI summary
An object identification device that can accurately identify a detected object is obtained. A radar inputs a radar confidence for an object in each of categories to a CPU; a camera inputs a camera confidence for an object in each of categories to the CPU; the CPU weighted-averages the respective confidences of the radar and the camera for each category so as to obtain the average confidence of each category; then, the category having a highest average confidence is identified as the kind of the object.