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? : support@ridgerun.com
Please visit our Main Website for the RidgeRun products and Professional Services