Development of a Hybrid Swarm Intelligence Convolution Neural Network for Object Detection and Tracking

Abstract

In this work, a Hybrid Particle Swarm Optimization Convolution Neural Network (CNN-HPSO) technique was developed to improve computational efficiency in terms of processing time and accuracy for object detection and tracking. The video datasets (MP4 and AVi video formats) used were pre-processed and then segmented. An Enhanced Particle Swarm Optimization (HPSO) was formulated from standard PSO and was applied to Convolution Neural Network (CNN) to form CNN-HPSO technique which was used for object tracking. The work was implemented with CNN-HPSO, CNN-PSO and CNN using MatLab R2016 software. The average results of CNN-HPSO, CNN-PSO and CNN on the videos with MP4 format yielded processing time, and accuracy of 165.89s, 97.08%; 179.52s, 94.25%, and 189.19s, 89.95%, respectively. For the videos in AVi format, CNN-HPSO, CNN-PSO and CNN produced similar average results with processing time and accuracy of 185.09s, 96.62%; 198.24s, 94.83%; and 216.59s, 91.30%, respectively. The findings revealed that the developed technique was highly computational efficient in terms of processing time and accuracy and can be used for solving other related optimization problems.