Reverse time ordered stroke context for air-writing recognition

Abstract

A novel real-time recognition system is proposed to recognize air-written characters without using any pen-starting-lift information. This pen-starting sign is commonly adopted in most of air-writing recognition systems for simplifying the complexity of trajectory matching but often results in inconvenience of usage for users. To tackle this problem, a novel reverse time ordered stroke context is proposed to represent an air-written trajectory in a backward way so that redundant starting-lift data can be effectively filtered out. Then the air-writing recognition problem can be formulated as a path finding problem which is easily solved by a stroke weighting scheme. Another two challenging problems, i.e., the multiplicity problem and the confusion problem also often happen in an air-writing recognition system. The first problem means a character is often written differently among different users. The second problem means different characters often own similar writing trajectory, e.g., {‘b’, ‘p’, ‘D’}. The two problems can be well tackled by introducing a new hierarchical classification scheme which constructs a three-layer structure to represent an air-writing character with different sampling rates. The first layer is designed for tackling the confusion problem by a grouping scheme. The second and third layers are used for dealing with the multiplicity problem of writing styles. All the alphabets (including lowercase, capital, and digital letters) are tested in this system and can be recognized in real time. Performance evaluation shows that the proposed solution achieves quite higher recognition accuracy (more than 94.7%) even though no starting gesture is required.