A vehicle-mounted video camera, which is one of low-cost off-the-shelf devices, can be used economically for pavement crack monitoring. The pavement frames obtained by the video camera can be merged to form a mosaic image, from which road distress information can be extracted. However, quality of crack detection in the frames is different from one another. The different level of crack detection quality should be considered for accurate construction of crack mosaic. This paper proposes a new pavement crack mosaicking method based on quality of crack detection in each frame. A convolutional neural network is suggested as a way to evaluate the quality of crack detection in the video frames. The proposed method showed a promising mosaicking performance compared to other existing methods.