Non-motorized travel data collection is an important part of the Intelligent Transportation System (ITS), which can be used for transportation planning, infrastructure design, traffic operation and management, etc. However, these data cannot be collected directly from commonly used detectors such as induction loops, sonar and microwaves. This paper proposes a novel low-cost WiFi-based monitoring system WiPedestrian, which can track and accurately class pedestrians and bicycles. The developed system includes four modules: Data Collection, Data Processing, Feature Extraction, and Pattern Classification. In the data processing module, the Received Signal Strength Indication (RSSI) filtering algorithm for low-speed pedestrian and bicycle hybrid traffic is proposed, which is effective to suppress the environmental noise caused by surrounding obstacles. And then, the travel time and instantaneous velocity characteristics are extracted by the WiPedestrain. Thirdly, the pattern classification module employs a recurrent neural network (RNN) model with long short time memory (LSTM). This study implemented WiPedestrian with off-the- shelf WiFi devices and conducted field data collection experiments at the campus from South China University of Technology, China. The results showed that the proposed RSSI filtering algorithm achieved better signal filtering effects in both static and dynamic environments. In addition, the classification accuracy of the system for mobile mode (walking and biking) is up to 97.92%. The proposed system is a feasible alternative for the data collection of non-motorized travel modes.