Data collected by Wireless Sensor Networks (WSNs) are inherently unreliable. Therefore, to ensure high data quality, secure monitoring, and reliable detection of interesting and critical events, outlier detection mechanisms are needed to be in place. The constraint nature of resources available in WSNs necessities that unlike traditional outlier detection techniques performed off-line, outliers to be identified in an online manner. This means that outliers in distributed streaming data should be detected in (near) real time with a high accuracy while maintaining the resource consumption of the WSN to a minimum. In this paper, we propose outlier detection techniques based on one-class quarter-sphere support vector machine meeting constraints and requirements of WSNs. To reduce the false alarm rate while increasing the detection rate and to enable collaborative outliers detection, we take advantage of spatial and temporal correlations that exist between sensor data. Experiments with both synthetic and real data show that our distributed and online outlier detection techniques achieve better detection accuracy and lower false alarm compared to an earlier distributed, batch outlier detection technique designed for WSNs.