TODO list

Here is a list of features that I would like to see in this package. The author of the package is open to cooperation and will be happy if someone is ready to join the development of the package.

1. General questions

• Checking the package for performance, searching for bottlenecks.
• Studying the issue of parallelization of the solution.
• Implementation of a sliding window.

2. Attributes and simple objects

• A check attribute that allows you to set limits on the sub_id of detection signs. This is convenient for unified work with recognition attributes that can produce a large number of different objects (CNN, Aruco).
• Integration of masked neural networks.
• A detailed study of the issue of connecting tensorflow api, writing a detailed guide to install it.
• Torch integration.
• Integration of face recognition (the ratio of the found face to the base of faces) as a attribute of information extraction.
• Expansion of the number of integrated recognition methods for key points. Writing your own logic over OpenCV that allows method data to find more than one object on the frame. Development of a metric for assessing the quality of recognition using the key point method
• Adding attributes working with contours.
• Replacement in the logic of the contour objects recognition system with masks. Adding orientation extraction attributes for objects with masks.
• Increase in the number of methods for clustering objects.
• PCL integration: segmentation of the observed scene into objects; determination of three-dimensional dimensions of the object, segmentation of the floor and walls (?).
• May be interesting for some tasks https://www.learnopencv.com/selective-search-for-object-detection-cpp-python/.
• Size filter and maxsize filter.

3. Tracking

• Learn how to run GOTURN.
• Add the ability to configure the object confirmation threshold, ie. an object is considered recognized when it has been recognized steadily for several frames.
• Display information about the history of moving the object to the topic.
• To develop a special kind of tracking for the soft mode, which allows tracking an object using an incomplete list of recognized features, while maintaining a high confidence coefficient.
• Integrate dlib tracking.

4. Relationships and complex objects

• Add dimension comparison relations, both in image coordinates and in 3D space.
• Add comparison relations of object orientation.
• Develop a soft mode for recognizing complex objects.