Automated Object Tracking on Resource-Constrained Devices

The patent segments the image processing pipeline into distinct stages: object detection using simple blob tracking, object tracking through frames, and classification using machine learning only when timing rules are satisfied. This segmentation allows the system to use computationally inexpensive methods for routine tasks while reserving resource-intensive machine learning for specific moments, thereby resolving the contradiction between accuracy and resource consumption.

The patent implements periodic classification by introducing timing rules that determine when machine learning-based classification should be performed. Instead of classifying every detected object in every frame, the system periodically applies classification based on timing conditions, reducing overall computational resource consumption while maintaining acceptable classification accuracy for the subset of objects processed.

The patent segments the tracking and classification functions into separate processing stages. Simple blob tracking is used for continuous object localization across frames, while complex classification is applied periodically to selected frames based on timing rules. This segmentation maintains tracking accuracy through continuous simple tracking while improving processing speed by limiting expensive operations to periodic intervals.

The patent applies partial action by performing classification only on a subset of detected objects and only when timing rules are satisfied, rather than classifying every object in every frame. This partial application of complex vision algorithms maintains adequate tracking accuracy for monitored objects while significantly improving overall processing speed and resource efficiency.

The patent segments detection and classification into separate stages performed by different methods. Simple blob tracking handles the detection and tracking functions with high processing speed, while machine learning-based classification is applied periodically to selected frames to provide accurate object type identification. This segmentation allows each method to excel at its designated task, resolving the contradiction between speed and accuracy.

The patent uses periodic classification based on timing rules to apply sophisticated machine learning analysis at intervals rather than continuously. This periodic application of accurate classification methods compensates for the limited accuracy of simple blob tracking while maintaining high processing speed during the intervals when only simple tracking is performed.