Lane-Splitting Motorcycle Detection Using Temporal Image Subtraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles and human drivers face challenges in detecting fast-moving motorcyclists lane-splitting between vehicles, especially in congested urban areas, due to occlusions and the difficulty in distinguishing them from surrounding traffic using existing sensing systems.
Innovation Solution
A system and method that includes a controller with imaging devices and sensors to analyze image streams, extract regions of interest, convert them to grayscale, and detect lane-splitting vehicles by subtracting reference images from current images to identify significant changes, triggering alerts and adjusting vehicle trajectory to avoid collisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional sensing systems are used to detect lane-splitting motorcyclists, then the system can detect stationary or slow-moving objects, but it fails to detect fast-moving motorcyclists changing lanes at higher relative speeds
Solution Approach 1:
The system dynamically adjusts its detection parameters based on the relative speed of detected objects. Fast-moving objects trigger different detection thresholds and processing algorithms compared to stationary objects, allowing the system to maintain measurement precision across a wide speed range by adapting to the dynamic characteristics of lane-splitting motorcyclists
Solution Approach 2:
The system adds temporal dimension to the detection by analyzing multiple frames over time and computing optical flow. This transforms the detection from a static spatial analysis to a spatiotemporal analysis, enabling the system to detect fast-moving motorcyclists by tracking their motion trajectory across multiple time points rather than relying solely on spatial presence
2Measurement precision
If standard image processing is used, then the system can process general traffic scenes, but it cannot reliably distinguish lane-splitting motorcyclists from surrounding traffic
Solution Approach 1:
The system segments the image processing task into distinct stages: initial object detection, optical flow computation, trajectory analysis, and classification. By dividing the complex task of distinguishing lane-splitting motorcyclists into manageable segments with specialized processing for each stage, the system achieves high distinction accuracy without requiring all processing complexity to be applied uniformly to every image
Solution Approach 2:
The system applies different processing quality and complexity to different regions of the image based on detected activity. Regions with high optical flow or unusual motion patterns receive more intensive processing and analysis, while static regions use standard processing. This local quality approach focuses computational resources on areas most likely to contain lane-splitting motorcyclists, improving distinction accuracy while managing overall system complexity
3Area of stationary object
If 360 degree sensing systems are deployed in autonomous vehicles, then coverage is improved, but objects or vehicles can still occlude the sensing system and prevent detection of lane-splitting motorcyclists
Solution Approach 1:
The system performs preliminary detection and tracking of potential lane-splitting motorcyclists before they are fully occluded or lost. By initiating detection early and maintaining tracking through predictive algorithms, the system can anticipate the presence of motorcyclists even when temporarily occluded by other vehicles, improving reliability by preparing for occlusion scenarios before they fully occur
Solution Approach 2:
The system uses feedback from multiple sensing angles and temporal sequences to compensate for occlusions. When one sensor or camera view is occluded, the system receives feedback from adjacent sensors and uses temporal prediction to maintain detection. This multi-source feedback mechanism ensures that occlusion of individual sensing elements does not compromise overall detection reliability
Data Source
AI summary
A controller receives images from a rearward-facing camera having an inter-lane region in the field of view thereof. A region of interest of the images is identified corresponding to the inter-lane region. The region of interest is converted to a grayscale image having a drastically reduced number of intensity levels. A difference image is determined between the grayscale image for each image relative to the grayscale image of a preceding image. If the non-zero pixels in the difference image exceed a threshold and the area of non-zero pixels in the difference images for the input images are increasing with time, a lane-splitting vehicle will be determined to be present and an alert may be generated and/or autonomous lane changes may be suppressed.