Time-of-flight range imaging cameras obtain range by producing amplitude modulated light and measuring the time taken for light to travel to the scene and back to the camera. Time-of-flight cameras require at least three raw measurements to calculate range. Raw frames are captured sequentially, and as such, motion in scenes during capture leads to inconsistent raw frame measurements and erroneous range calculations. Motion error constrains Time-of-flight cameras to non-dynamic scenes and limits their potential applications. The Time-of-flight bidirectional Kalman filter method is a state-of-the-art method known to reduce error due to transverse motion in cameras operating with a single modulation frequency. The method works by treating raw frames as a noisy time series and running the Kalman filter on it to produce a range estimation at every raw frame. The Kalman filter is then reapplied to the data in reverse to produce another set of range estimations, and a composite range is selected from the two set of range estimations. A number of commercial timeof-flight cameras, such as the Microsoft Kinect V2, use multiple modulation frequencies. In this thesis, we adapt the bidirectional Kalman filter method to multi-frequency operated cameras by having the prediction component of the Kalman filter take into account the change in amplitude and phase shift caused by the change in frequency. The amplitude component of the prediction is performed linearly by multiplication, while the phase shift component of the prediction is performed using the ratio of the modulation frequencies. The phase shift prediction across modulation frequencies requires the phase to be unwrapped. The unwrapping is performed between modulation frequencies by selecting the number of phase wraps that best predicts the two following raw frames. Finally, to ensure correct composite phase selection, an alternative method for selecting the composite phase is proposed for the adapted bidirectional Kalman filter. We perform quantitative and qualitative experiments to test the proposed method. In the quantitative experiment, the proposed method produces less error than the classical Discrete Fourier Transform approach in 70% of tested instances. The qualitative experiment shows that the proposed method significantly reduces motion blur.