Pedestrian Detection Unit with Dynamic Criterion Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition devices struggle to accurately detect pedestrians, especially those who appear different from typical pedestrians, leading to lower detection rates and false negatives or false positives.
Innovation Solution
An image recognition device with an image acquisition unit, moving object detection unit, and pedestrian detection unit that relaxes the detection criterion for pedestrians in regions where moving objects are detected, using a combination of optical flow and pattern matching methods to enhance detection accuracy and prevent false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a constant detection criterion is used for pedestrian detection regardless of moving objects, then false positives are reduced, but pedestrians in various situations (e.g., pushing strollers) are missed
Solution Approach 1:
The detection criterion is made dynamic by adjusting it based on the presence of moving objects. When a moving object is detected, the system switches to a relaxed detection criterion in that region, allowing detection of atypical pedestrians. This dynamic adjustment resolves the contradiction between maintaining strict criteria for accuracy and relaxing criteria for adaptability.
Solution Approach 2:
The system applies different detection criteria to different regions of the image based on local conditions. In regions where moving objects are detected, a relaxed criterion is applied specifically to that local area, while other regions maintain the standard criterion. This local differentiation allows the system to adapt to various pedestrian situations without compromising overall detection reliability.
2Adaptability or versatility
If the detection criterion is relaxed to detect atypical pedestrians, then detection coverage improves, but false positives increase
Solution Approach 1:
The relaxed detection criterion is applied locally only in regions where moving objects are detected, rather than globally across the entire image. This localized application ensures that the relaxed criterion benefits atypical pedestrian detection in relevant areas while maintaining strict criteria in other areas, thus preventing unnecessary false positives.
Solution Approach 2:
The image is effectively segmented into regions with and without moving objects, with different detection criteria applied to each segment. This segmentation allows the system to relax criteria only where necessary (in moving object regions) while maintaining strict criteria elsewhere, balancing detection coverage and false positive rate.
3Reliability
If strict detection criteria are applied, then false positives are minimized, but detection rate for atypical pedestrians decreases
Solution Approach 1:
The detection criterion dynamically adapts based on the presence of moving objects. In regions with moving objects, the criterion becomes more lenient to capture atypical pedestrians, while maintaining high precision in regions without moving objects. This dynamic behavior resolves the contradiction between precision and detection rate.
Data Source
AI summary
An image recognition device includes an image acquisition unit, a moving object detection unit, and a pedestrian detection unit. The image acquisition unit acquires consecutive captured images of a surrounding of a vehicle from a camera mounted to the vehicle. The moving object detection unit calculates a moving amount based on consecutive captured images in time series acquired by the image acquisition unit to detect a moving object from the captured images. The pedestrian detection unit detects a pedestrian using an identifier for identifying a pedestrian from the captured images acquired by the image acquisition unit. The pedestrian detection unit relaxes a detection criterion for detecting a pedestrian using the identifier in a region of a captured image where the moving object is detected.