Today, we can witness wireless sensor networks (WSNs) in action almost everywhere. Their applications are ubiquitous covering environment, medical care, military, surveillance, etc. While the potential benefits of WSNs are real and significant, there remains two major challenges to fully realize this potential: big data collection and limited sensor energy. To overcome these problems, filtering techniques over data routed to the sink should be used in such a way that they do not discard useful information. In this paper, we propose a new filtering technique dedicated to periodic sensor applications. The first filter is applied at the sensor nodes and aims to reduce their raw data based on the Pearson coefficient metric. The second filter is applied at intermediate nodes, called aggregators. It uses Knearest neighbor clustering algorithm in order to eliminate data redundancy collected by neighboring nodes. In order to evaluate our technique, experiments on real telosB sensors have been conducted while the results showed significant energy savings and high accurate data collection compared to existing approaches.