Controlling System Volume Using YOLO Deep Learning Algorithm Through Hand Gesture Recognition

Abstract

Dynamic hand gesture recognition is challenging in the real world because the gestures performed varies from person to person making it difficult to classify and detect the appropriate gesture. Also, the system must classify the gesture as soon as it detects because a lag in the detection and classification will result in inappropriate results. In fact, a negative lag is favorable as it will provide instantaneous feedback to the user. The model used for this purpose is the YOLO model. Majority of the recent work in yolo has focused on classifying pre-segmented video clips, and some progress has also been made on joint detection and recognition of actions in complex video sequences. In the process of object detection first step is the detection of the object which is moving at different position. The detection part can be done by various methods firstly by the apparent movement of images i.e. Optical Flow secondly by Spatiotemporal Salient Region Detection, Yolo, 3D CNN, Inception etc. After the detection part is done of the moving object the next step can be of classifying the objects on the basis of their structure, texture or motion whether it is a human or non-living object. After the classification done, the third step is to control the system volume speed using fps (frames per second). We address these challenges with Yolo model that detects the hand movement and based on the direction of that movement we increase or decrease the system volume.

Keywords: Classification, CNN, Machine learning, Optical flow.

View PDF