Object Detection Device Preventing Incorrect Grouping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems based on range images often incorrectly group multiple objects at the same distance as a single object, leading to inadequate vehicle control actions, such as failing to decelerate properly for pedestrians or incorrectly identifying preceding vehicles.
Innovation Solution
The system employs a method that generates range images from gray-scaled images using stereo cameras, where valid and invalid range data are evaluated, and objects are grouped based on luminance differences in gray-scaled images to prevent incorrect grouping by dividing regions according to vertical edges and merging sub-regions appropriately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If grouping is performed based on valid and invalid range data in a range image, then object detection can be performed using distance information, but correct grouping of multiple objects at the same distance cannot be achieved
Solution Approach 1:
The patent divides the detection process into two independent segmentation stages: first segmenting the image into valid/invalid range data regions based on distance, then segmenting into object groups based on luminance characteristics. This two-stage segmentation allows precise separation of multiple objects at the same distance by using luminance differences as an additional segmentation criterion, resolving the contradiction between detection capability and grouping accuracy.
Solution Approach 2:
The patent transitions from one-dimensional distance-based grouping to two-dimensional grouping by incorporating luminance information as a second dimension. By evaluating luminance values and luminance difference values in addition to range data, the system creates a multi-dimensional detection space that enables accurate distinction between objects at the same distance, thereby improving grouping accuracy without sacrificing detection productivity.
2Device complexity
If only range data is used for object detection, then distance-based object identification is possible, but objects at the same distance cannot be distinguished
Solution Approach 1:
The patent merges two different data types (range data from stereo vision and luminance data from gray-scaled images) into a unified detection framework. By combining distance information with luminance information, the system achieves reliable object identification for objects at the same distance, improving reliability while maintaining reasonable system complexity through integrated processing.
Solution Approach 2:
The patent introduces luminance difference values as an intermediary parameter that mediates between range data and object identification. This intermediary enables the system to distinguish objects at the same distance by evaluating luminance characteristics, thereby improving object identification reliability without significantly increasing device complexity.
3Speed
If grouping is performed based on range image data, then detection speed is maintained, but incorrect grouping of multiple objects occurs
Solution Approach 1:
The patent performs preliminary evaluation of luminance characteristics and luminance difference values before final object grouping. By pre-calculating and storing luminance information, the system enables rapid comparison and grouping operations, maintaining detection speed while improving grouping precision through advance preparation of discriminative features.
Solution Approach 2:
The patent replaces complex mechanical or computational grouping mechanisms with a simpler luminance-based comparison system. By using luminance difference values as the primary grouping criterion, the system achieves precise object separation through straightforward numerical comparison, maintaining high detection speed while improving grouping precision.
Data Source
Figure 1
Figure 2
Figure 3(a)~4
AI summary
To provide an object detection device and an object detection method in which incorrect grouping based on a range image can be prevented, processing is implemented including: generating a range image based on a pair of gray-scaled images (S603); grouping adjacent regions having range data representing close distances in the range image (S604); calculating differences between the luminance values of pixels in a vertical line below a ground plane above which objects are imaged and the luminance values of pixels in a vertical line above the ground plane in the gray-scaled images and dividing the group region into laterally separated sub-regions (S605); and determining whether or not there is a section without vertical edge in either of the left and right areas of each of sub-regions made based on the gray-scaled images and merging sub-regions whose areas including a section without vertical edge are next to each other (S607).