Road sign content prediction and search in smart data management for training machine learning model

A method for training a machine learning model to recognize road signs, the method comprising:

• maintaining image data including video having a sequence of frames;
• detecting a road sign in one of the frames;
• cropping at least one of the frames based on the detected road sign to produce a copped image that includes the road sign;
• executing a visual attribute prediction model to determine a plurality of probabilities of a presence of predefined visual attributes present in the cropped image;
• querying a knowledge graph reasoner to access a plurality of rules related to the predefined visual attributes;
• validating the visual attribute prediction model by removing at least one of the probabilities of presence of predefined visual attributes in the cropped image based on the rules;
• and
• outputting at least one sign candidate from a plurality of stored sign templates, wherein the at least one sign candidate matches the road sign within the cropped image based on the validated visual attribute prediction model.

A system for training a machine learning model to recognize road signs, the system comprising:

• a user interface;
• a storage configured to maintain (i) image data of a plurality of images, (ii) a list of predefined visual attributes found in images of road signs, and (iii) a plurality of stored sign template images;
• and
• a processor in communication with the storage and the user interface, the processor programmed to:
• receive a cropped image of a road sign detected in an environment surrounding a vehicle,
• execute a visual attribute prediction model to determine whether one or more of the predefined visual attributes are present in the cropped image, wherein the visual attribute prediction model is configured to assign a probability of presence to each of the one or more predefined visual attributes,
• query a knowledge graph reasoner to access a plurality of rules relating to the one or more predefined visual attributes present in the cropped image,
• validate the visual attribute prediction model by removing the probability of presence of at least one of the predefined visual attributes present in the cropped image based on the rules, and
• output a sign template image from the plurality of stored sign template images that matches the road sign within the cropped image based on the validated visual attribute prediction model.