Object Identification with DNN Retraining for Unseen Item Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object-identification systems face challenges in accurately detecting and identifying items due to factors like lighting conditions, shadows, and obstructed views, leading to inconsistent results.
Innovation Solution
A machine-learning-assisted object-identification system using a deep neural network (DNN) that incorporates image and depth sensors, with a controller for image processing, object detection, and human interaction for retraining, enabling continuous improvement through human-supplied information and multi-authentication processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If video monitoring and image processing techniques are used to identify items, then automation of the purchase and checkout process is achieved, but detection accuracy deteriorates due to lighting conditions, shadows, and obstructed views
Solution Approach 1:
The system implements feedback by collecting instances where the deep neural network fails to recognize objects, obtaining human-provided labeling information for these cases, and using this feedback to retrain the network. This continuous feedback loop progressively improves detection accuracy while maintaining automation.
Solution Approach 2:
The system performs self-service by automatically identifying and collecting challenging detection cases, soliciting human input only when needed, and autonomously retraining the deep neural network using the collected data. This minimizes ongoing human intervention while continuously improving performance.
2Speed
If a deep neural network is used for object recognition, then recognition speed is improved, but recognition accuracy deteriorates for previously unseen or difficult objects
Solution Approach 1:
The system performs preliminary action by proactively identifying detection failures and collecting training data in advance. When the deep neural network fails to recognize an object, the system immediately captures the image, obtains human labeling, and adds it to the training set before encountering similar difficult cases again, thereby preparing the network for future recognition tasks.
Solution Approach 2:
The system ensures continuity of useful action by maintaining constant operation of the deep neural network for rapid recognition while simultaneously and continuously improving it through periodic retraining on newly collected difficult cases. This allows the system to maintain high speed while progressively enhancing accuracy.
3Measurement precision
If human input is collected for retraining the deep neural network, then recognition accuracy is improved, but system complexity increases
Solution Approach 1:
The system extracts only the essential and necessary human input - specifically, labeling information for cases where the deep neural network fails to recognize objects. By taking out only this critical information rather than requiring comprehensive human annotation of all images, the system improves accuracy while minimizing the added complexity.
Solution Approach 2:
The system discards cases where the deep neural network successfully recognizes objects, requiring no human input. It recovers and utilizes only the failed recognition cases for human labeling and retraining. This selective approach maximizes accuracy improvement while minimizing system complexity by avoiding unnecessary human intervention.
Data Source
AI summary
A system and method of identifying and tracking objects comprises registering an identity of a person who visits an area designated for holding objects, capturing an image of the area designated for holding objects, submitting a version of the image to a deep neural network trained to detect and recognize objects in images like those objects held in the designated area, detecting an object in the version of the image, associating the registered identity of the person with the detected object, retraining the deep neural network using the version of the image if the deep neural network is unable to recognize the detected object, and tracking a location of the detected object while the detected object is in the area designated for holding objects.