Stereo Vision for Dimension Estimation

We designed a system for inferring package volumetric dimensions in collaboration with a Canadian transportation company, Rogue Specialty Transport. When a client hires a transportation company to ship their goods, the main component used to calculate cost of delivery is the volume of the package. Currently, the standard is to hire personnel to traverse the warehousing facilities prior to shipment, and measure the dimensions using a pen-and-paper system. This is costly, prone to human error, and is very time consuming.

In response to this process breakdown, we designed stereoscopic system that calculates the dimensions given four corners. That is, the system calculates the real-world coordinate of the image points, and then calculates the real-world distance between these points. This is made possible since the disparity can be determined across the two stereo images.

The technical workflow is as follows (see the video for an example):

  1. Live stereoscopic images are streamed through the interface.
  2. When the object of interest is properly in both frames, the user clicks "Capture" to grab the frame and perform rectification.
  3. The user then clicks on an origin corner, and three other corners with which to measure.
  4. The system calculates the corresponding real-world coordinates using the stereoscopic disparity, and outputs the real-world distance between the points.

Some realistic examples are provided below.

Example 1 showing estimated dimension of 108x139x159

 Actual dimensions: 101cm x 122cm x 164cm

 Example 2 showing estimated dimension of 19 x 23 x 45

 Actual dimensions: 20cm x 25cm x 45cm

Example 4 showing estimated dimension of 114 x 122 x 90

 Actual dimensions: 101cm x 122cm x 88cm


Related people


Alexander WongDavid A. Clausi


Robert AmelardFarnoud Kazemzadeh


Christian Scharfenberger

Related research areas

Computer Vision