Real-time Human Tracking System using Histogram Intersection Distance in Firefly Optimization Based Particle Filter

Abstract

: Real-time human tracking in a video have numerous applications. For security and surveillance application, the tracking system with PTZ (Pan, Tilt, and Zoom) camera is expected to track an object correctly regardless of the object orientation. Numerous studies reported that Particle Filter (PF) is reliable for color object tracking. However, the PF algorithm still suffers from impoverishment and degeneration in the resampling process. These problems can be resolved by combining the PF with Firefly Optimization (FO) in the resampling process. This research proposes the use of Histogram Intersection distance to build a likelihood function in PF to achieve real-time implementation. The Firefly Optimization Algorithm-based Particle Filter (FOAPF) with Histogram Intersection distance was compared to FOAPF with Bhattacharyya distance, resulting in lower RMSE (Root Mean Square Error) in tracking TB datasets. The result shows that when the Histogram Intersection distance was implemented, a faster average time of 1.8 ms was achieved than 1.9 ms when using Bhattacharyya distance. It shows the time result slightly different. The FOAPF with Histogram Intersection distance results in the TB datasets perform a low RMSE of 4.96 and 12.07, and private datasets show a low RMSE of 16.92 and 8.80, with the real-time implementation of 30 FPS and 50 particles. The comparison presents the successful implementation of the proposed method as a tracker to enhance human movement tracking with real-time implementation.