SYSTEM AND METHOD FOR SPRAY PAINT IMAGE SYNTHESIS IN SURVEILLANCE CAMERA ANOMALY DETECTION

The present invention discloses a system and method for detecting anomaly in a surveillance camera. The method allows the construction of an anomaly surveillance image dataset for training a surveillance camera anomaly detection deep neural network model. The method includes a spray paint images synthesizer that is used for constructing the dataset comprises the steps of: estimating a spray density map to represent the thickness of spray paint; preparing a Gaussian kernel with variance varied at each image location depending on the spray density; synthesizing the blurring effect given by the spray paint by mathematically convolving a surveillance image with the spatial-varying Gaussian function; estimating a spray alpha map to decide how strong the color appears on the image at each location; performing an element-wise weighted sum of the spray paint blur image and the spray paint color image with the weight depending on the spray alpha map; randomizing the design parameters to generate spray paint images of different spray paint patterns, thickness, and colors.