# This object tracking solutions utilize the "Triangle Similarity" method. # In brief, the triangle similarity takes an object (marker) with a known # width. The object is placed at some distance from the camera. Preferably # the same camera to be used for detection and tracking. # # A photo/image is taken of the object using the same camera to be used for # object detection. We then measure the apparent width in pixels. Using # this image as a fixed reference point, if the camera moves away from the # object, the number of pixels measuring the object's width decreases. If # the camera moves closer to the object, the number of pixels increases. # One can calculate the distance with pretty good accuracy. The higher the # quality of the reference image, the better the accuracy. These attributes # include: good lighting, good color contrast, accurate distance, as close to # 90% camera angle as possible. In some cases, camera calibration and # focal length maybe required. However I found using the piCam,pinhole or # fish-eye distortion wasn't an issue. # # Solution Approach # This solution uses a reference or calibration image of the object to # track. The object/marker width is determined by the number of pixels. # Once this has been established, the main loop does the following: # - looks for an object with similar contours as the reference object # - if found, target identified # - determines target width in pixels # - calculates distance based on a % difference from the # reference/calibration image # - locate target center # - display/print target details # - for Raspberry Pi, get CPU temperature as well. # # This version previews the target, paints the target's boarders and provides # target data on the preview screen. This an excellent method of viewing # and debugging the code. Also included were performance stats for the # various functions. If using for robotics or autonomous mode, all the # above can be commented/removed for maximum performance. Through testing, # with good target LIGHTING, I was achieving 30+ FPS. (Raspberry Pi 3, # piCam ver2, Multitheaded Camera streaming feed, Python 3.6.