Speedup the Multi-camera Video-Surveillance System for Elder Falling Detection

Nowadays, all countries have to face the growing populations of elders. For most elders, unpredictable falling accidents may occur at the corner of stairs or a long corridor due to body functional decay. If we delay to rescue a falling elder who is likely fainting, more serious consequent injury may happen. Traditional secure or video surveillance systems need someone to monitor a centralized screen continuously, or need an elder to wear sensors to detect accidental falling signals, which explicitly require higher costs of care staffs or cause inconvenience for an elder.In this work, we propose a human-shape-based falling detection algorithm and implement this algorithm in a multi-camera video surveillance system. The algorithm uses multiple cameras to fetch the images from different regions required to monitor. It then uses a falling-pattern recognition approach to determine if an accidental falling has occurred. If yes, the system will send short messages to someone needs to alert.In addition, we propose a multi-video-stream processing algorithm to speedup the throughput for the video surveillance system having multiple cameras. We partition the workloads of each video-surveillance streaming into four tasks: image fetch, image processing, human-shape generation, and pattern recognition. Each task will be handled by a forked thread. When the system receives multiple video streams from cameras, there are four simultaneous threads executed for different tasks. The objective of this algorithm is to exploit large thread-level-parallelism among those video-stream operations, and apply pipelining technique to execute these threads.  All above algorithms have been implemented in a real-world environment for functionality proof. We also measure the system performance after multi-streaming speedup. The results show that the throughput can be improved by about 2.12 times for a four-camera surveillance system.