Aiming at the problem that in traditional WiFi-pedestrian track reckoning（ PDR） localization algorithm

there are some errors in the initial position provided by WiFi

and there are accumulative errors in PDR positioning itself. These two factors lead to the decline of positioning accuracy. An activity recognition aided PDR indoor localization algorithm based on convolutional neural network（ CNN） is proposed. Several access points（ AP）are arranged in the experimental area. The AP signal intensity threshold RSSI

θ

obtained in the feature area is used as a trigger. When a mobile device detects a signal（ RSSI） greater than RSSI

θ

then it triggers this algorithm t

o determine whether the user is active in the feature area and whether it matches the current feature point activity

if so

The position is corrected to the position of the current feature point

otherwise

continue PDR. The experimental results show that compared with the traditional WiFi-PDR localization algorithm

the precision of the feature points is increased by 33. 5 %. And the overall positioning precision is improved by 24. 6 %.