top of page

Deep Learning Action Recognition: Automatic Mixed Martial Arts Scoring

Action recognition is an active research field in Deep Learning. This is a more challenging task than normal instance detection since action patterns emerge not only from spatial neighboring pixels, but temporal ones too.

Such complexity opens a new range of smart and dynamic systems. The following examples show how action recognition can be used to automate expensive and repetitive tasks:

  • Quality control in factory production lines

  • Sports fairness and rule compliance

  • Active CCTV surveillance

RidgeRun R&D is creating a system to automatically score Mixed Martial Arts fights. The system should not only detect and classify the fighters but identify hit distance, target and effectiveness, takedowns, and overall ground control. In general, the project involves:

  • Acquire and prepare 100+ fighting hours

  • Carefully label and augment individual actions

  • Train and optimize hyperparameters

  • Use case development

  • Validation and deployment

Are you looking to apply video analytics to automate a step in your process? Please

Learn more about this project and all the possibilities in our developer's wiki :

Any Questions? :

Please visit our Main Website for the RidgeRun products and Professional Services

bottom of page