Camera-based Assessment of Gendered Toy Preference in Free-Play Parent-Child Interactions

This repository focuses on detecting joint attention and mutual gaze in parent-child interactions during free play. We leverage advanced computer vision techniques to achieve this.

Head Tracking

Head Detection

• Method: We utilize the head detection module from LAEO-Net++.

Head Tracking

• Method: Head tracking is accomplished using the DeepSORT method, available in the DeepSORT repository.

Interpolation

To handle missing detections and smoothen the tracking results, run the interpolation script: ```bash headtracking/interpolate.py ```

Object Detection

For detecting objects in the scene, [add details or methods used if any].

Classification

Gaze Attention Heatmap

• Method: Gaze attention heatmaps are generated using the method proposed in Detecting Attended Visual Targets in Video.

Visual Focus of Attention (VFOA)

We integrate the gaze attention heatmap with the detected regions to determine if an object is being attended to by either the parent or child. A threshold of 80 is used to make this determination.

Mutual/Joint Attention

By combining the views from both the parent and child, we derive mutual or joint attention insights.

Evaluation

For evaluating the results and obtaining performance metrics, run the Cohen's Kappa evaluation script: ```bash evaluate/cohen-kappa.py ```

---

Thank you for exploring this repository. Contributions and suggestions are welcome! Feel free to open issues or submit pull requests.