Camera Target Detection via Image Cell Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current camera systems for motor vehicles face challenges in precisely detecting target objects, particularly when features associated with the ground are erroneously classified as ground features, leading to incomplete and imprecise detection of objects like pedestrians.
Innovation Solution
A method that divides the image into image cells, classifies them as ground or target cells, and combines adjacent cells to form target clusters, using optical flow vectors correlated with the vehicle's movement to differentiate features, and adjusts for features near the ground by defining regions of interest and correcting cluster positions based on intensity and deviation vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If characteristic features are classified based on optical flow vectors to differentiate ground features from target features, then the classification accuracy is improved, but features near the ground are erroneously classified as ground features leading to incomplete detection
Solution Approach 1:
The image is divided into multiple image cells that are then grouped into clusters. By segmenting the image into cells and forming clusters, the patent enables separate handling of ground features and target features within the same spatial region, allowing features near the ground to be correctly associated with target objects rather than erroneously classified as pure ground features.
Solution Approach 2:
Adjacent image cells are combined to form target clusters that represent complete target objects. This merging process integrates both ground features and target features that belong to the same object, ensuring that features near the ground are not lost but properly associated with their parent target object through the cluster structure.
2Device complexity
If only adjacent target features are combined for detection, then the grouping process is simplified, but ground features associated with target objects are excluded leading to incomplete detection
Solution Approach 1:
The patent segments the image into image cells and classifies each cell as either a ground cell or a target cell based on optical flow analysis. This segmentation allows the system to identify which cells contain target features and which contain ground features, enabling precise control over the grouping process.
Solution Approach 2:
The patent merges adjacent target cells and ground cells to form unified target clusters. This combining process ensures that ground features associated with target objects are included in the same cluster as the target features, creating complete and accurate target object representations without requiring complex selective grouping logic.
3Reliability
If the entire image is processed for target detection, then detection coverage is maximized, but computational effort increases reducing real-time processing capability
Solution Approach 1:
The patent divides the image into multiple smaller image cells, which can be processed independently and in parallel. This segmentation reduces the computational complexity of processing the entire image while maintaining complete coverage, as each cell can be evaluated quickly for ground or target classification.
Solution Approach 2:
Adjacent image cells are merged into clusters that represent meaningful target objects. This merging reduces the total number of elements that need to be processed and reported, as multiple cells are combined into single target cluster representations, thereby reducing overall computational effort while preserving detection coverage.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables precise detection of target objects and accurate determination of bounding boxes, reducing computational effort and enabling real-time detection, thereby improving the accuracy and efficiency of object recognition in vehicle camera systems.
Implementation Method 1
a camera (3) disposed on the motor vehicle (1) and capturing an environmental region (6) of the motor vehicle (1)
Implementation Method 2
the optical flow method can be used by determining the so-called optical flow vector to each characteristic feature based on two images
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for detecting a target object (5a, 5b, 5c) in an environmental region of a camera based on an image (7) of the environmental region provided by means of the camera by determining a plurality of characteristic features in the image (7), wherein it is differentiated between ground features and target features; by dividing at least a partial region (12) of the image (7) in a plurality of image cells (13) and examining to each image cell (13) whether the image cell (13) includes a ground feature or a target feature; by classifying those image cells (13), which include a target feature, as target cells (13b) and those image cells (13), which include a ground feature, as ground cells (13a) and by detecting the target object (5a, 5b, 5c) by combining a plurality of adjacent image cells (13) to a target cluster considering both the target cells (13b) and the ground cells (13a).