Risk prediction method

A risk prediction method executed by a computer of a risk predictor using a convolutional neural network, the method comprising:

• taking an input image by an in-vehicle camera installed on a vehicle;
• acquiring, by the computer using the convolutional neural network, an input image taken by an in-vehicle camera installed on a vehicle;
• making the convolutional neural network estimate a plurality of risk areas, including a first risk area having a detected first moving object, and a second risk area not having a detected moving object detected from a side of the second risk area toward the vehicle as a second moving object, and a degree of risk for each risk area of the plurality of risk areas as a feature of each risk area, each risk area being in the acquired input image,
• the first risk area having a possibility that the detected first moving object in the first risk area may move into a travelling path of the vehicle, and that the detected first moving object in the first risk area may collide with the vehicle, and
• the second risk area having a possibility that the second moving object may appear by a direction from the second risk area into a travelling path of the vehicle, and that the second moving object may collide with the vehicle, the possibility that the second moving object may appear from the second risk area being based on a degree of similarity of the acquired input image to a partial area of an input image taken immediately before, a previous situation where a previous moving object did appear from a previous second risk area not having a detected moving object;
• displaying the input image taken by the in-vehicle camera installed on the vehicle on a display installed in the vehicle;
• and
• superimposing, by the convolutional neural network, a likelihood map for each risk area of the plurality of risk areas on the input image being displayed on the display installed in the vehicle,
• wherein the likelihood map for the first risk area includes a first irregularly-shaped boundary line that surrounds the detected first moving object, and
• wherein the likelihood map for the second risk area includes a partial first irregularly-shaped boundary line associated with the second risk area to show a direction of the second moving object may appear from an other side of the second risk area opposite the vehicle.