Human Detection Using Occupancy Grid Maps for Occluded Scenes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human detection systems in images struggle to accurately identify individuals in crowded environments where objects partially or fully occlude each other, making it difficult to apply foreground blob extraction techniques effectively.
Innovation Solution
The use of occupancy grid maps combined with a deep learning framework, where an occupancy grid generator creates a grid representing image areas, a human renderer generates training data by adding rendered human characters, and a scene-specific detector trains using these data to accurately detect humans in specific camera scenes, with an image augmenter overlaying detected humans for visualization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If foreground blob extraction or visible human shape identification is used, then the detection method is simple, but detection accuracy deteriorates in crowded environments with occlusions
Solution Approach 1:
The patent transforms the detection problem from 2D image space to 3D occupancy grid space. By projecting detected human blobs onto a 3D occupancy grid and analyzing occupancy patterns across multiple depth layers, the system can distinguish occluded humans from background objects even when their 2D projections overlap. This dimensional transformation enables accurate detection in crowded scenes where traditional 2D methods fail.
2Ease of manufacture
If traditional human detection methods are used, then the system is easy to implement, but reliability deteriorates when objects partially or fully occlude each other
Solution Approach 1:
The patent introduces an occupancy grid as an intermediary data structure between image processing and human detection. The occupancy grid accumulates depth information and occupancy patterns from multiple image processing results, serving as a mediator that resolves ambiguities caused by occlusions. This intermediary layer enables reliable detection by providing additional spatial context without requiring complex changes to the overall system architecture.
3Productivity
If foreground blob extraction is used, then processing is fast, but detection precision worsens in crowded environments
Solution Approach 1:
The patent performs preliminary processing by generating occupancy grids from multiple image processing results before final human detection. The occupancy grid pre-computes depth information and occupancy patterns, which are then used to guide the final detection process. This preliminary action separates the fast blob extraction step from the precision detection step, maintaining processing speed while improving accuracy through pre-computed spatial context.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for detecting humans in images using occupancy grid maps. The methods, systems, and apparatus include actions of obtaining an image of a scene without people, generating a reference occupancy grid from the image, generating pairs of training images with humans rendered and corresponding training occupancy grids based on the occupancy grid and the image, training a scene-specific human detector with the pairs of training images with humans rendered and corresponding training occupancy grids, generating a sample occupancy grid from a sample image using the scene-specific human detector, and augmenting the sample image using the sample occupancy grid.