Gating Camera Object Identification for Overlapping Depth Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Monocular cameras struggle to accurately separate overlapping objects at different distances, and existing TOF cameras are costly and require extensive learning data for effective object identification.
Innovation Solution
Employ a gating camera to divide the field of view into multiple ranges, capturing images at varying exposure times to generate separate object images, and use a processing device with multiple classifiers or scaling to enhance identification accuracy and reduce learning costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a monocular camera is used for object detection, then the device cost is reduced, but the ability to separate overlapping objects at different distances is lost
Solution Approach 1:
The camera divides the field of view in the depth direction into multiple ranges and captures images for each range separately by changing exposure timings. This segmentation allows objects at different distances to be captured in separate images, enabling the classifier to identify each object without interference from overlapping objects, thus resolving the contradiction between cost and separation capability
Solution Approach 2:
The patent introduces depth range as an additional dimension by dividing the field of view into multiple depth ranges. Instead of capturing a single 2D image, the system captures multiple images corresponding to different depth ranges, adding the depth dimension to the image data and enabling separation of overlapping objects
2Measurement precision
If a TOF camera is used to acquire depth information, then object separation capability is improved, but device cost increases
Solution Approach 1:
Instead of using an expensive TOF camera, the patent uses a conventional camera to capture multiple images at different exposure timings. These multiple 2D images serve as copies that collectively provide depth information, replacing the need for a costly TOF sensor while achieving the same object separation capability
Solution Approach 2:
The patent replaces the optical-mechanical TOF system with a temporal-mechanical approach using exposure timing control. Instead of measuring time of flight optically, the system uses controlled exposure at different timings to separate objects by depth, substituting a complex optical measurement system with a simpler temporal control mechanism
3Reliability
If extensive learning data is used for classifier optimization, then object identification rate is improved, but learning cost increases
Solution Approach 1:
The patent applies different classifiers optimized for specific depth ranges rather than using a single classifier for all ranges. Each classifier is trained on locally relevant data from its corresponding depth range, improving identification accuracy for objects in that specific range while reducing the total learning data required compared to training one universal classifier on all data
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
Improves object identification rates by separating overlapping objects and reduces learning costs through data augmentation, enabling high-precision object detection with a combination of algorithms tailored to near and far distances.
Implementation Method 1
A TOF (Time Of Flight) camera is configured to emit infrared light by means of a light-emitting device, to measure the time of flight up to the time point at which the reflected light returns to the image sensor, and to convert the time of flight into distance information in the form of an image
Data Source
AI summary
An object identification system includes a camera and a processing device. The processing device includes a classifier subjected to machine learning based on the output image of the camera so as to allow it to identify an object. A gating camera divides a field of view in the depth direction into multiple ranges, and captures an image while changing the time difference between light projection and exposure for each range. The classifier is subjected to machine learning using multiple images IMG1 through IMGN generated by the gating camera as the learning data.