Particle Filter-Based Single Shot MultiBox Detector for Predicting Human Movement

Peneliti : Carmadi Machbub, Lenni Yulianti, Pranoto Hidaya Rusmin
Student: Devira Anggi Maharani

Tracking moving objects is currently a focus of attention for researchers. .. Correctly tracking moving objects is quite a challenging problem, especially when the objects experience occlusions, lighting changes, unexpected movements, and assume different movement poses. To improve the accuracy of the moving object detector, we propose SSD (Single Shot Multi-box Detector) selain PF (Particle Filterto provide moving human prediction. Performance evaluation is done by comparison with HOG-SVM as a detector. The proposed system has been successfully tested on two videos. SSD-based PF with 100 particles performed well, with an RMSE of 7.44 and 91.24 effective particles. The results show that adding SSD in the measurement process can improve the performance of PF for tracking moving humans. The results also show that the proposed method is successfully applied in combination with specific color detection to track specific human objects.