It is designed to provide a harness for object detection algorithms, to be used with both test-datasets and live-SRVC-capture.

It uses the ROS messaging structure.

Current status: compiled, checked in, not yet tested. a couple of function implementations still to write.

TODO

• write dummy feature generation and object detection stubs, in both python and C++.
• write the '.launch' xml file to run the group of processes. http://www.ros.org/wiki/roslaunch
• write get_category_index function (in junior_manager.cpp)
• write the SVRC_Scores::add function (in harness_collector.cpp)
• write the evaluate_overlap function (in harness_collector.cpp)
• write the add and save functions for the SVRC_Scores structure
• Check that all process will guarentee to de-allocate all the heap memory that they allocate.
• To achieve some level of sanity testing, the dummy object detection stubs could be allowed to look at the 'right' answer, and return this answer with probability p. The resulting ROC and SRVC Scores could then be checked to see if they correspond with what is predicted in that case.
• Write a 'senior_manager' that will integrate the intermediate data structures returned by the object detectors (about the multiple views) and then supply this (integrated) data structure back to the object detectors for them to provide a hypothesis. (This last bit hasn't really been talked about too much. It means writing another service for the Object Detector program, that will input an integrated vector of votes/bag of words, and output the hypothesis.)

description of programs

There are five programs that comprise the package:

Issues