Thanks to technological advancements, we now have a number of options for obtaining traffic data. Yet, the precision with which various technologies measure drivers' actual speeds varies greatly. It is a widespread complaint among transportation researchers and professionals that they lack information of device accuracy. This study employs video data image processing alongside the Euclidean distance approach to estimate vehicle speeds from a variety of perspectives. To begin, the proposed approach applies preprocessing to the frames we retrieved from the video data in order to reduce the shadow impact. Then, we use a Gaussian Mixture Model (GMM) to pull out the foreground from the background. The next step is to apply a median filter on the resulting foreground. Using Haar Wavelet decomposition for feature extraction, appearance-based hypothesis verification ensures the correctness of hypotheses based on their outward appearance. After that, FILM-CNN is used to train the model which outperforms k-Means and CNN methods.