摘要 |
A method of improving performance of video analytics in a field-deployed camera system (100), the method comprising: receiving image data representing multiple images of a scene, the multiple images including representations of multiple objects, wherein a first set of the objects are members of an object class and a second set of the objects are not; analysing the image data to determine whether the multiple objects are either members or non-members of the object class, wherein a mistaken determination is made; generating mistake metadata 1014 based on acknowledgement by a user 104 of the mistaken determination; and improving automatically the video analytics performance based on the mistake metadata so as to reduce a number of future mistaken determinations. The mistake metadata may be added to an N-step cascaded object classifier, becoming step N+1. The improved classifier may be validated using the data which initially led to the mistakes or errors in classification. |