Real-Time Pedestrian Pose Estimation for Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing advanced driver assistance systems (ADAS) face challenges in real-time detection of vulnerable road users, such as pedestrians and cyclists, especially in crowded scenes or when individuals are small or far away, leading to compromised accuracy and increased false detections.
Innovation Solution
A novel lightweight person pose estimation method and system that employs a top-down approach combining a lightweight backbone infrastructure with a confidence pre-processing procedure. This system generates boundary boxes around detected individuals and uses trackers to optimize detection accuracy, reduce false positives, and recover missed detections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If top-down pose estimation methods are used to detect small or far-away pedestrians, then detection accuracy for small objects improves, but inference time increases proportionally with the number of people in the image
Solution Approach 1:
The system performs preliminary object detection to identify all pedestrians in the scene before executing pose estimation. This preliminary detection step creates a filtered set of candidate regions, allowing the subsequent pose estimation to focus only on detected individuals rather than processing the entire image, thus reducing inference time while maintaining accuracy for small or far-away pedestrians
Solution Approach 2:
The pose estimation process is segmented into two distinct phases: first, object detection to locate all pedestrians and generate bounding boxes; second, pose estimation applied only to the detected regions. This segmentation allows the system to optimize each stage independently, using efficient detection algorithms to identify candidates and then applying accurate pose estimation only where needed
2Productivity
If bottom-up pose estimation methods are used for real-time detection, then inference time is reduced and real-time performance is achieved, but detection accuracy drops for small or far-away pedestrians
Solution Approach 1:
The system performs preliminary object detection to identify all pedestrians in the scene before executing pose estimation. This preliminary detection step creates a filtered set of candidate regions, allowing the subsequent pose estimation to focus only on detected individuals rather than processing the entire image, thus reducing inference time while maintaining accuracy for small or far-away pedestrians
Solution Approach 2:
The system applies different processing qualities to different regions of the image based on detection results. High-resolution pose estimation is applied only to detected pedestrian regions, while undetected regions receive no processing. This local quality approach concentrates computational resources where they are most needed, improving accuracy for small objects without proportionally increasing overall computation time
3Measurement precision
If high-resolution processing is applied to detect small or far-away pedestrians, then detection accuracy improves, but computational complexity increases
Solution Approach 1:
The image processing is segmented into detection regions and non-detection regions. High-resolution processing is applied only to the small subset of the image containing detected pedestrians, while the majority of the image receives minimal or no processing. This segmentation dramatically reduces the total computational complexity while maintaining high detection accuracy for small or far-away pedestrians in the detected regions
Solution Approach 2:
The system applies different processing qualities to different regions of the image based on detection results. High-resolution pose estimation is applied only to detected pedestrian regions, while undetected regions receive no processing. This local quality approach concentrates computational resources where they are most needed, improving accuracy for small objects without proportionally increasing overall computation time
Data Source
AI summary
The present invention discloses a system for detecting poses of people around a vehicle. An electronic sensor such as a camera is associated with a vehicle for the generation of consecutive frames within a video. The system generates a boundary box around a person within each frame and optimizes a tracker for each person in consecutive frames. The system detects people on the road through pose estimation. The system performs a confidence pre-processing procedure before sending the trackers to a pose estimator to extract their poses.