Version: 1.7.0

Requirements

Hardware

  • The required HW specifications depend mostly on the implementation of the application. Most applications connecting to our cloud-based service work on all mobile devices of the past 5 years without performance-issues.
  • The camera should be able to record video at 640x480px or more at 30 frames per second or more, otherwise quality of our service is decreased.

Image Dimensions

  • The optimal image dimensions for analysing lying down exercises are 640x480 pixels (landscape).
  • The optimal image dimensions for analysing standing exercises are 480x640 pixels (portrait).
  • However, both portrait and landscape can always be sent, but analysis performance might degrade with respect to the optimal case.
  • We recommend to retain the aspect ratio when scaling down the images and then scaling the returned points up proportionally. An example of how to do this can be found within the swift-getting started section.
  • Generally, a 4 : 3 aspect ratio of images yields the best results.

Please note that images with a higher resolution generally lead to a better analysis performance but increase the latency. And vice versa images with a lower resolution improve the latency but decrease the analysis performance.

Image Encoding

  • We recommend an image size of 30kB to ensure a high enough frame rate that leads to satisfying motion analysis results.
  • The image data must be encoded either as JPEG or PNG. To reach the desired image size, we recommend to tune the compression rate to keep the image size close to the desired size. Based on our experience, a good initial compression quality is 60%. An example of how to encode an image can be found within the swift-getting started section.

Image Orientation

  • The orientation of the image must be upright, meaning gravity showing downwards. Especially for landscape images, it must be made sure that they are oriented correctly.
  • The image should not be mirrored. Some mobile phones might return mirrored images depending on the respective hardware, so it must be made sure that those are flipped accordingly before sending to the API.

Please see following image as an example.

results matching ""

    No results matching ""