CPC G06V 10/765 (2022.01) [G06V 10/993 (2022.01); G06V 20/52 (2022.01); G08B 21/02 (2013.01); G06V 10/82 (2022.01)] | 5 Claims |
1. A method for detecting objects thrown from height, comprising the following steps of:
setting up a video surveillance area for capturing a target video image in real time;
detecting all moving objects within the target video image using a background modeling algorithm;
tracking the detected moving objects using Kalman filtering to obtain trajectory characteristics and related parameters of the moving objects, including trajectory curve trend, object movement acceleration, intersection over union of previous and current frames of the object, object shape and pixel size changes, then all trajectories being analyzed and filtered;
identifying whether the moving objects are objects thrown from height through rule-based identification of objects thrown from height, setting an outer contour line of the building and a special position line, so that once the trajectory of a moving object crosses the contour line from inside to outside, or crosses the special position line from top to bottom, the object is identified as an object thrown from height;
performing classification using a Long Short-Term Memory (LSTM) network classification model, including inputting trajectory curve trend, object movement acceleration, intersection over union of previous and current frames of the object, object shape and pixel size changes as feature data into the LSTM, to obtain a classification result and determine whether misinformation occurs, wherein an input layer of the LSTM network includes extracted feature data including trajectory curve trend, object moving acceleration, intersection over union of previous and current frames of the object, object shape and pixel size changes, a number of neurons in a first LSTM hidden layer is 128, a number of neurons in a second LSTM hidden layer is 32, and a number of neurons in a final output layer is 1, which represents a probability of objects thrown from height; and
pushing an alarm to the monitoring center if the classification result of the LSTM classification model does not indicate misinformation.
|