Accurate box localization for automatic tagging system using physical ground truth

A method for object detection, the method comprises:

• obtaining one or more inference sensed information units of an environment of a vehicle;
• and
• detecting one or more inference objects captured within the one or more sensed information units;
• wherein the detecting is executed, at least in part, by one or more object detection machine learning processes that were trained by tagged sensed information units;
• wherein the tagged sensed information units were generated by an automatic tagging process that comprises:
• (g) obtaining sensed information units;
• (h) obtaining distance information regarding distances between a training vehicle and objects captured in the sensed information units;
• (i) identifying one or more dimension-indicative-properties related to the objects;
• the one or more dimension-indicative-properties related to the objects differ from one or more actual dimensions related to the objects;
• (j) calculating the one or more actual dimensions related to the objects, based on the one or more dimension-indicative-properties related to the objects;
• (k) determining one or more sensed information units dimensions related to the objects, based on the distance information and the one or more actual dimensions related to the objects;
• and
• (l) generating tags associated with the objects, wherein the tags are indicative of at least one of (i) locations related to the objects within the sensed information units, and (ii) the one or more sensed information units dimensions related to the objects;
• and
• wherein at least steps (c)-(f) are executed by one or more automatic tagging machine learning processes.

A non-transitory computer readable medium for automatic tagging, the non-transitory computer readable medium stores instructions for:

• (a) obtaining sensed information units;
• (b) obtaining distance information regarding distances between a vehicle and objects captured in the sensed information units;
• (c) identifying one or more dimension-indicative-properties related to the objects;
• the one or more dimension-indicative-properties related to the objects differ from one or more actual dimensions related to the objects;
• (d) calculating the one or more actual dimensions related to the objects, based on the one or more dimension-indicative-properties related to the objects;
• (e) determining one or more sensed information units dimensions related to the objects, based on the distance information and the one or more actual dimensions related to the objects;
• and
• (f) generating tags associated with the objects, wherein the tags are indicative of at least one of (i) locations related to the objects within the sensed information units, and (ii) the one or more sensed information units dimensions related to the objects;
• and
• wherein at least steps (c)-(f) are executed by using a machine learning process.

A vehicle processing circuit that is configured to:

• obtain bounding shapes that bound objects, the objects were captured in a sensed information unit;
• generate a cropped image for each bounding shape;
• segmenting each cropped image to a cropped image background and a cropped image foreground;
• and
• generate, for each cropped image, an updated bounding shape that are indicative of dimensions of the cropped image foreground of the cropped image, to provide updated bounding shapes.