Abstract: A method for motion detection and a circuit system are provided. In the method, when any moving object within a scene is sensed, continuous frame images are obtained. Each of the frame images is divided into multiple regions. Brightness and chromaticity of each region are calculated frame by frame, and are compared with a background model to obtain a difference. Accordingly, a movement event in each of the frame images can be identified. When any movement event is recognized, a region of interest (ROI) covering one or more of the regions where the movement event is detected is established. An intelligent model created by a neural network algorithm is used to calculate a trust score for a possibility of having the moving object in each of the frame images. The moving object can be detected by comparing the trust score with a determination threshold.