Autism has symptoms can hardly be recognized in the early stages of the disease, and it affects the child's mental health on the long term. Autism can be identified by parents monitoring to the child and diagnosed by psychiatrists using an international standard checklist. The checklist questions should be answered by the parent and psychiatrist to determine the risk level of autism (high, medium, or low risk). It is hard for parents to monitor more than 20 child's behaviours at the same time regardless lack of accuracy for answering on most of these questions. We propose a system for monitoring autistic child behaviours by analysing accelerometer data collected from wearable mobile device. The behaviours are recognized by using a novel algorithm called DTWDir that based on calculating displacement and direction between two signals. DTWDir is evaluated by comparing it to KNN, classical Dynamic Time Warping (DTW), and One Dollar Recognition ($1) algorithms. The results show that DTWDir accuracy is higher than the others.