Trailer Loading Visibility Using Camera-Based Object and Fill-Level ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to provide real-time analytics for object detection and fill-level determination during loading and unloading processes, leading to inefficiencies and inaccuracies, such as unnecessary door openings and closures, and reliance on manual monitoring or resource-intensive LiDAR sensors.
Innovation Solution
A computerized system employs machine learning models, including an object-detection ML model and a fill-level ML model, to analyze video data from existing cameras to automatically detect, track, and classify objects, and determine fill-levels in storage compartments, reducing computational resources and manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual monitoring is used to track loading/unloading processes, then operational control can be maintained, but labor costs increase and accuracy decreases
Solution Approach 1:
The patent replaces manual mechanical monitoring with an automated computer vision system using cameras and machine learning models. The system captures images of storage compartments, processes them through ML models to detect objects and determine fill-levels, thereby eliminating the need for manual monitoring while improving accuracy and reducing labor costs.
2Measurement precision
If LiDAR sensors are used for object detection and fill-level determination, then measurement precision improves, but computational resources and costs increase
Solution Approach 1:
The patent uses standard cameras instead of expensive LiDAR sensors to capture images for analysis. While individual camera frames are simple and inexpensive, the system processes them efficiently through optimized machine learning models, achieving comparable detection accuracy at lower hardware and computational costs.
Solution Approach 2:
The patent extracts only the essential features needed for object detection and fill-level determination from camera images, rather than processing complete 3D spatial data like LiDAR. This extraction approach focuses computational resources on relevant visual features, reducing overall computational burden while maintaining detection precision.
3Productivity
If real-time analytics are generated during loading/unloading, then productivity improves, but processing time and computational load increase
Solution Approach 1:
The patent pre-trains machine learning models offline using labeled datasets before deployment. During actual loading/unloading operations, the pre-trained models perform rapid inference on new images, enabling real-time analytics generation without the computational overhead of training during operational periods. This separation of training and inference phases improves real-time processing efficiency.
Data Source
AI summary
Provided are embodiments for providing analytics indicative of object detection or fill-level detection at or near real-time based on video data captured during an unloading or loading process. A computerized system may detect and classify, using an object-detection machine learning (ML) model, an object based on the video data. A computerized system may further determine, using a fill-level ML model, a fill-level of the storage compartment based on a comparison of edges of the storage compartment to a total dimension corresponding to the edge. In this manner, the various implementations described herein provide a technique for computing systems employing image processing and machine learning techniques to a video data stream to generate analytics associated with the unloading or loading process at or near real-time.