CASSANDRA: audio-video sensor fusion for aggression detection
AVSS '07 Proceedings of the 2007 IEEE Conference on Advanced Video and Signal Based Surveillance
| Hi-index | 0.00 |
The paper describes a surveillance system of cameras installed at lamppost of a military area. The surveillance system has been designed to detect unwanted visitors or suspicious behaviors. The area is composed of streets, building blocks and surrounded by gates and water. The video recordings are monitored by operators at regular times. Next some operators inspect the area by car. To compute the probability of unobserved intrusion we used (Hidden) Markov Models. We also investigated the weak spots. We considered three intrusion scenarios of suspicious individuals, traveling by car, by boat and by foot. The military area has been split up in different areas/zones, partly covered by cameras or by the inspection team. The probability of detection in some forbidden areas has been extracted from the surveillance system and from the time tables of the inspection operators. The transition probabilities between areas are computed using data from analysis of video recordings and based on interviews with operators.