Pedestrian Detection via Object Visibility Map Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pedestrian detection methods face challenges in accurately detecting pedestrians, especially when they are blocked, due to poor detection performance and increased training data complexity.
Innovation Solution
A pedestrian detection method that involves obtaining an image, performing feature extraction to generate a basic feature map, determining a proposal for a pedestrian, creating an object visibility map to highlight visible pedestrian features, and fusing this map with the basic feature map to enhance detection accuracy, thereby improving the detection of blocked pedestrians without increasing training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional pedestrian detection methods are used, then the detection process is simple, but the detection accuracy for blocked pedestrians is poor
Solution Approach 1:
The patent segments the pedestrian detection process into multiple stages: generating an object visibility map to identify visible and occluded regions, then selectively processing these regions differently. The visibility map divides the detection task into visible pedestrian parts (processed with standard detection) and occluded parts (processed with specialized handling), thereby improving overall accuracy without uniformly increasing complexity across the entire system.
Solution Approach 2:
The patent introduces an object visibility map as an intermediary element between the input image and the final detection result. This visibility map serves as a mediator that guides the detection process by highlighting occluded regions, allowing the system to focus computational resources where they are most needed while maintaining a relatively simple overall architecture.
2Measurement precision
If more training data is used to improve detection accuracy, then the detection performance improves, but the training data complexity and requirements increase
Solution Approach 1:
The system generates the object visibility map automatically from the input image through processing by the same neural network model used for detection. This self-generated guidance information eliminates the need for separate annotated visibility maps or additional training data, allowing the model to improve its own detection capability through self-service rather than requiring external data resources.
Solution Approach 2:
The patent changes the parameter representation by generating an object visibility map that transforms the original image into a guided representation highlighting occluded regions. This parameter transformation allows the model to leverage existing training data more effectively by providing explicit guidance about where to focus attention, thereby improving accuracy without requiring additional training samples.
3Productivity
If the detection method is simplified, then the processing speed increases, but the detection accuracy for occluded pedestrians decreases
Solution Approach 1:
The patent applies partial action by generating the object visibility map only for regions where occlusion is likely to occur, rather than processing the entire image uniformly. The visibility map provides selective guidance only where needed, maintaining fast processing for clear regions while applying enhanced detection only to occluded areas, thus balancing speed and accuracy.
Solution Approach 2:
The patent implements local quality by applying different processing strategies to different regions of the image based on the object visibility map. Visible regions are processed with standard detection methods for speed, while occluded regions identified in the visibility map receive specialized processing attention, thereby optimizing the trade-off between processing speed and detection accuracy in different local areas.
Data Source
AI summary
This application relates to the field of artificial intelligence and the field of computer vision. The method includes performing feature extraction on an image to obtain a basic feature map of the image, and determining a proposal of a region possibly including a pedestrian in the image. The basic feature map of the image is then processed to obtain an object visibility map in which a response to a pedestrian visible part is greater than a response to a pedestrian blocked part and a background part. The method further performs weighted summation processing on the object visibility map and the basic feature map to obtain an enhanced feature map of the image, and determines, based on the proposal of the image and the enhanced feature map of the image, a bounding box including a pedestrian in the image and a confidence level of the bounding box including the pedestrian.