In this paper, we propose a robust activity recognition approach for smart healthcare using body sensors and deep convolutional neural network (CNN). We analyze signals from different body sensors for healthcare, such as ECG, magnetometer, accelerometer, and gyroscope sensors. After extracting salient features from the sensor data based on Gaussian kernel-based principal component analysis and Z-score normalization, a deep activity CNN is trained based on the features. Finally, the trained deep CNN is used for recognizing the activities in testing data. The proposed approach is applied to a publicly available standard data set and then compared with other conventional approaches. The experimental results show that the proposed approach is superior than others, indicating the robustness of the approach to be adopted for cognitive assistance in body sensor-based smart healthcare systems.