Decision Rules based on hand location and area change

[周游]

Hallo everyone,

Based on hand location detection after background subtraction I have constructed a series of decision rules to recognize hand gestures. I think that it can be used to modify neural network or to train a deep one. 

Decision Rules have two parts.

First part classifies gestures in 4 classes (Horizontal, Vertical, Cental, REJ)

Second part classifies gestures in each class

The following is the code of first part decision rules:

if(max_x-min_x < POSSIBLE_MOVEMENT_THRESHOLD){ // not possible to be a horizontal movement

        if(max_y-min_y < POSSIBLE_MOVEMENT_THRESHOLD){ // not possible to be a vertical movement

            return JudgeCentalMovement(a_s,va_s);

        }else{ // possible to be a vertical movement

            if(max_y-min_y < CERTAIN_MOVEMENT_THRESHOLD){

                string s=JudgeVerticalMovement(vy_s)+ " OR " + JudgeCentalMovement(a_s,va_s);

                return s;

            }else{

                return JudgeVerticalMovement(vy_s);

            }

        }

    }else{// possible to be a horizontal movement

        if(max_y-min_y < POSSIBLE_MOVEMENT_THRESHOLD){ // not possible to be a vertical movement

            if(max_x-min_x < CERTAIN_MOVEMENT_THRESHOLD){

                string s = JudgeHorizontalMovement(vx_s)+ " OR " + JudgeCentalMovement(a_s,va_s);

                return s;

            }else{

                return JudgeHorizontalMovement(vx_s);

            }

        }else{ // possible to be a vertical movement

            if(max_x-min_x > CERTAIN_MOVEMENT_THRESHOLD){

                if(max_y-min_y > CERTAIN_MOVEMENT_THRESHOLD)

                    return "REJ";

                else{

                    return JudgeHorizontalMovement(vx_s);

                }

            }else{

                if(max_y-min_y > CERTAIN_MOVEMENT_THRESHOLD)

                    return JudgeVerticalMovement(vy_s);

                else{

                    string s=JudgeHorizontalMovement(vx_s)+" OR ";

                    s=s+JudgeVerticalMovement(vy_s)+" OR ";

                    s=s+JudgeCentalMovement(a_s,va_s);

                    return s;

                }

            }

        }

    }

The attachment contains the location and area sequence of each gesture and the result of decision rules. (479 bag files)

Best regards

--
Zhou You

周 游