Video Object Annotation Propagation for Dense Part Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manually annotating pictures for video data in manufacturing processes is inefficient, particularly in applications like bowl feeders, where objects are densely populated, leading to significant production losses due to downtime.
Innovation Solution
A method involving a point propagation algorithm to annotate and track objects in a video, using a segmentation model to generate bounding boxes, and train an object detection model, which reduces manual annotation effort and enhances tracking efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation of video frames is performed to train object detection models, then detection accuracy is improved, but annotation time and human cost increase significantly
Solution Approach 1:
The system performs preliminary actions by detecting objects in only the first frame of a video sequence and using that information to generate annotations for subsequent frames through tracking algorithms. This preliminary detection and tracking approach eliminates the need for manual annotation of every frame, reducing annotation time while maintaining detection accuracy through the use of trained models on the annotated first frame and propagated annotations.
2Measurement precision
If manual annotation of densely populated objects is performed, then tracking accuracy is improved, but annotation complexity and difficulty increase
Solution Approach 1:
The system replaces the mechanical manual annotation process with automated computer vision algorithms. Object detection models and tracking algorithms automatically identify and annotate densely populated objects in video frames, eliminating the need for manual annotation while maintaining high tracking accuracy. The automation handles the complexity of densely populated objects through algorithmic object recognition and tracking.
3Reliability
If traditional object detection methods are used in manufacturing processes, then production monitoring is achieved, but production losses occur due to system downtime
Solution Approach 1:
The system enables continuous monitoring of manufacturing processes by processing video streams in real-time without requiring manual intervention or system downtime for annotation. The automated detection and tracking algorithms continuously analyze video frames to monitor part flow, detect abnormalities, and maintain production efficiency without interruption, ensuring both reliability and productivity.
Data Source
AI summary
Systems and methods for annotating and tracking objects in a video are described herein. The methods include operating at least one processor to: receive, from at least one image device proximal to the manufacturing device, a sequence of frames of a video showing the plurality of parts within the manufacturing device; receive at least one annotated frame having labelling of a subset of parts of the plurality of parts in a plurality of frames of the video, the annotated frame being video annotation data; apply the video annotation data as input to a propagation algorithm to annotate an additional subset of parts of the plurality of parts within the frames of the video, the additional annotated frames being additional video annotation data; apply a segmentation model to the additional video annotation data to generate image segmentation masks of each of the parts, the image segmentation masks being trained segmentation model output data; and apply an object detection model to the trained segmentation model output tracking data to get a fine-tuned object detection model to detect and track the parts.