Silhouette quality quantification for gait sequence analysis and recognition

Most gait analysis and recognition algorithms are designed based on silhouette data. Their recognition performances may therefore be impaired by the poor quality silhouettes extracted from outdoor environments. In this paper, a silhouette quality quantification (SQQ) method is proposed to assess the quality of silhouette sequence. SQQ analyzes the sequence quality based on 1D foreground-sum signal modeling and signal processing technique. As an immediate application of SQQ, a general enhancement framework namely silhouette quality weighting (SQW) is designed toward improving most of the current gait recognition algorithms by taking into consideration sequence quality. The experiments are performed on the USF HumanID gait dataset v1.7 (with 71 subjects). Investigation using the SQQ criterion has revealed the baseline algorithm's mechanism of silhouette quality consideration. Two instantiations of the SQW algorithm based on gait energy image (GEI) are implemented. Improved recognition performances compared to the original GEI and baseline methods are obtained, which verifies the effectiveness of the proposed SQQ, SQW methods.