Pedestrian Detection Using 3D Depth Maps and Cascade Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pedestrian detection systems in vehicles face challenges in achieving high detection rates with low false positives in cluttered environments while maintaining real-time processing speed, particularly in urban areas where pedestrians may suddenly appear in the vehicle's path.
Innovation Solution
A computer-implemented method using a combination of 3D depth maps and appearance information to detect pedestrians, employing a cascade of classifiers tuned for multiple depth bands and trained on filtered data, including contour and gradient features, and utilizing deep learning convolutional networks for improved performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sparse stereo processing is used to generate putative pedestrian regions, then detection speed is improved, but detection accuracy deteriorates with 40% probability of detection and 1.65 false positives per image frame
Solution Approach 1:
The patent segments the detection process into multiple stages: sparse stereo processing generates initial putative pedestrian regions, then a cascade of classifiers (including HOG-based and appearance-based classifiers) progressively refines these regions. This multi-stage segmentation allows the system to maintain high processing speed in early stages while achieving high detection accuracy in later stages, resolving the contradiction between speed and accuracy.
Solution Approach 2:
The patent transitions from 2D image processing to 3D depth map processing by incorporating disparity information from stereo cameras. This dimensional change allows the system to leverage depth cues for more accurate pedestrian detection while maintaining real-time performance through efficient 3D processing algorithms.
2Reliability
If dense stereo processing is used for entire image, then detection accuracy is improved, but processing speed deteriorates due to real-time computation limitations
Solution Approach 1:
The patent segments the image processing into sparse stereo processing for initial detection and selective dense processing only for putative pedestrian regions. This avoids the computational burden of processing the entire image densely while maintaining high detection accuracy where it matters most.
Solution Approach 2:
The patent applies different processing qualities to different regions: sparse stereo processing is applied to the entire image for speed, while dense stereo processing and multiple classifier stages are applied locally only to putative pedestrian regions of interest. This local quality approach maintains accuracy where needed while preserving overall processing speed.
3Reliability
If temporal information is used to improve per-frame detection performance, then detection accuracy is improved, but latency of alerting the driver increases
Solution Approach 1:
The patent applies temporal information selectively rather than comprehensively. The cascade of classifiers processes each frame independently with optional temporal smoothing, applying temporal constraints only when they provide benefit without significantly increasing latency. This partial application of temporal information improves accuracy while minimizing time loss.
4Productivity
If monocular vision system with focus of attention mechanism is used, then processing speed is improved, but detection accuracy in cluttered environments deteriorates
Solution Approach 1:
The patent enhances the monocular focus of attention mechanism by incorporating 3D depth information from stereo processing. This dimensional enhancement allows the system to maintain high processing speed through selective region processing while improving detection accuracy in cluttered environments through depth-based discrimination.
Data Source
AI summary
A computer implemented method for detecting the presence of one or more pedestrians in the vicinity of the vehicle is disclosed. Imagery of a scene is received from at least one image capturing device. A depth map is derived from the imagery. A plurality of pedestrian candidate regions of interest (ROIs) is detected from the depth map by matching each of the plurality of ROIs with a 3D human shape model. At least a portion of the candidate ROIs is classified by employing a cascade of classifiers tuned for a plurality of depth bands and trained on a filtered representation of data within the portion of candidate ROIs to determine whether at least one pedestrian is proximal to the vehicle.