Human Activity Recognition (HAR) is a challenging task due to the complexity of human motions and the variability of datasets. The wide adoption of wearable devices and the incorporation of high-grade motion sensors and biosensors in these devices increased the number of available data that could be employed in sensor-based HAR. In this paper, we propose a multi-modal deep convolutional neural network capable of recognizing different activities using accelerometer data from several body positions. We compare the performance of our proposed system with existing DCNN model architectures. Our experiments on two public datasets for HAR demonstrated that our approach surpassed the performance of both single-position and simple multi-position DCNN models. This study provides valuable insights for the development of efficient edge-AI systems for activity recognition on resource-constrained embedded devices.