Performance analysis of real-time face detection system based on stream data mining frameworks
Nikolay Lvovich Kazanskiy, Vladimir I. Protsenko, П. Г. Серафимович · Procedia Engineering · 2017
This work describes performance analysis results of two face detection systems based on Apache Storm and IBM InfoSphere Streams frameworks. Profiling was evaluated on image sequences of four different sizes: 100 x 100, 640 x 640, 1920 x 1080, and 4096 x 3112. Face detection was performed by OpenCV cascade classifier. Experiment was held under five CentOS nodes cluster. It was investigated that system based on Apache Storm was able to operate in real-time at 24 frames per second on used hardware configuration. Apache Storm was more scalable and demonstrated advantage in throughput over its counterpart. Experiment helped to reveal configuration parameters of frameworks that played a major role in face detection task on image sequences.