Noise removal from surveillance camera image by means of AI-based object recognition

A surveillance camera comprising:

• a camera comprising a plurality of infrared light emitting diodes (IR LEDs);
• and
• at least one processor configured to:
• recognize an object through a deep learning-based object recognition algorithm from an image obtained through the camera,
• partition the image into M×N blocks, each block comprising a plurality of pixels,
• determine at least one target IR LED corresponding to coordinate information of the object among the plurality of IR LEDS,
• determine an object block comprising at least one block that includes the object,
• determine, for each of the M×N blocks, a respective first average brightness based on brightness values of a plurality of pixels included in each block,
• calculate a second average brightness of the object block by summing the first average brightness of only the at least one block included in the object block and dividing the sum by a number of the at least one block included in the object block,
• compare the second average brightness with a predetermined reference brightness, and
• in response to the second average brightness being lower than the predetermined reference brightness, control a brightness of the at least one target IR LED such that the second average brightness of the object block reaches the predetermined reference brightness,
• wherein the predetermined reference brightness refers to a minimum intensity of light required for object recognition in a low-light environment and is obtained through model training by defining illumination information and image data as input data and IR LED brightness information corresponding to a target object recognition rate as output data.

A control method for a surveillance camera comprising:

• partitioning an image acquired from a camera comprising a plurality of infrared light emitting diodes (IR LEDs) corresponding to a plurality of illumination areas into M×N blocks, each block comprising a plurality of pixels;
• recognizing an object through a deep learning-based object recognition algorithm;
• determining an object block comprising at least one block that includes the object;
• determining, for each of the M×N blocks, a respective first average brightness based on brightness values of a plurality of pixels included in each block;
• calculating a second average brightness of the object block by summing the first average brightness of only the at least one block included in the object block and dividing the sum by a number of the at least one block included in the object block;
• comparing the second average brightness with a predetermined reference brightness;
• and
• in response to the second average brightness being lower than the predetermined reference brightness, controlling a brightness of at least one target IR LED, among the plurality of IR LEDs, corresponding to an illumination area that includes the object block such that the second average brightness reaches the predetermined reference brightness,
• wherein the predetermined reference brightness refers to a minimum intensity of light required for object recognition in a low-light environment and is obtained through model training by defining illumination information and image data as input data and IR LED brightness information corresponding to a target object recognition rate as output data.