Video-Based Large Vehicle Detection Without Camera Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems struggle to accurately determine the size of vehicles, particularly large vehicles, due to variations in size within vehicle types and distortions caused by camera viewing angles, and often require costly and impractical camera calibration.
Innovation Solution
A system that uses an electronic processor to analyze vehicle trajectories and positions within video frames, determining vehicle length by computing a line segment parallel to the trajectory and comparing it to average car lengths, using a size multiplier to identify large vehicles without camera calibration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera calibration is performed to accurately determine vehicle size, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent introduces an intermediary reference vehicle (a standard-sized vehicle) as a mediator between the camera system and the target large vehicle. By detecting the reference vehicle's bounding box and using its known actual size, the system calculates a scaling factor that compensates for perspective distortion without requiring camera calibration. This intermediary enables accurate size measurement while avoiding complex calibration procedures.
Solution Approach 2:
The system performs self-calibration by automatically detecting reference vehicles in the video stream and computing the scaling factor from their bounding boxes. No manual camera calibration is needed - the system serves itself by using naturally occurring reference vehicles in the scene to establish the relationship between pixel measurements and real-world dimensions.
2Productivity
If bounding box size is used to determine vehicle type, then detection speed is improved, but measurement precision deteriorates due to viewing angle distortions
Solution Approach 1:
The patent transforms the raw bounding box dimensions into a corrected vehicle size estimate by applying a scaling factor derived from reference vehicle detection. Instead of directly using the bounding box width or height, the system computes: estimated_actual_size = bounding_box_dimension × scaling_factor. This parameter transformation compensates for perspective distortion while maintaining the computational efficiency of bounding box-based detection.
3Adaptability or versatility
If classification-based vehicle detection is used to handle size variations, then adaptability is improved, but measurement precision worsens due to generalization errors
Solution Approach 1:
The patent segments the vehicle detection problem into two independent components: (1) object detection and classification using bounding boxes, and (2) size estimation using the scaling factor method. By separating these functions, the system can accurately classify vehicle types while independently measuring their actual sizes without the generalization errors that plague classification-based size estimation.
Data Source
AI summary
Examples provide a system for detecting large vehicles. The system includes an electronic processor configured to obtain video data from an image sensor, determine a bounding box bounding a detected vehicle in a plurality of frames of the video data, determine a trajectory of the detected vehicle based on changes in location of the bounding box in the plurality of frames, and determine a position of the detected vehicle within at least one frame of the plurality of frames. Based on the trajectory of the detected vehicle, the bounding box, and the position of the detected vehicle within the at least one frame, the electronic processor determines a length of the detected vehicle, and determines whether the length is greater than a threshold. In response to the length of the detected vehicle being greater than a threshold, the electronic processor makes a determination that a large vehicle is detected.