Application of headcount estimation to animal counts based on crowd patch learning
Application of NEC’s unique headcount estimation tool, that learns crowd image patches by a convolutional neural network (CNN) where overlapping of people is reproduced by image synthesis to estimate numbers of people per patch, enables estimation of the number and density of animals with a broad scope of activity.
- Learns crowd images that show placement patterns of crowds overlapping enabling reliable estimations
- Estimates can be made by an existing camera with a shallow depression angle, making a specially placed camera necessary
- Is able to estimate numbers and density of people and animals even under congested conditions
- Can make estimates from just one image and is also applicable to low frame rate images