Gaze-Guided Work Estimation for Multi-Object Contact Recognition

The patent introduces an attention map as an intermediary element that mediates between the raw video input and the behavior recognition output. The attention map highlights regions of interest in the video frame, allowing the system to focus on relevant objects while ignoring irrelevant ones. This intermediary mechanism enables automated work estimation while preventing erroneous recognition caused by multiple objects in the field of view.

The patent applies local quality by making different regions of the video frame have different recognition weights. Instead of treating all detected objects equally, the system assigns higher importance to objects within the attention map's highlighted regions and lower importance to objects outside these regions. This local differentiation allows accurate behavior recognition even when multiple objects are present in the video.

The attention map serves as a filtering intermediary between object detection and work target identification. It allows the system to maintain comprehensive object detection capability while reliably distinguishing actual work targets from background objects or irrelevant items by highlighting only the regions that require attention.

The system applies different quality weights to different detected objects based on their location within the attention map. Objects falling within the attention map's highlighted regions are treated as high-priority work targets, while objects outside these regions are treated as low-priority or non-work targets, thereby improving recognition reliability without reducing adaptability.

The attention map introduces a moderate level of complexity as an intermediary processing step between simple object detection and final behavior recognition. While this adds some computational overhead, it dramatically improves estimation precision by enabling the system to distinguish relevant from irrelevant objects without requiring complex mechanical or structural modifications.

The patent replaces complex mechanical or manual verification systems with an attention-based computational mechanism. Instead of using elaborate physical sensors or manual confirmation processes to identify work targets, the system uses attention maps generated through image processing algorithms, achieving high precision with relatively simple computational infrastructure.