In the past few decades, the number of elderly people who prefer to live independently is significantly increasing among the elderly people due to the issues of privacy invasion and elderly care cost. Device-free non-privacy invasive activity recognition is preferred for long-term monitoring. Thus, we propose a deep learning classification method for elderly activities using binary sensors (PIR sensor and door sensor). In particular, we present a Deep Convolutional Neural Network (DCNN) classification approach for detecting four basic activity classes which represent the basic human activities in a home monitoring environment, namely: Bed-to-Toilet, Eating, Meal-Preparation, and Relax. A real-world long-term annotated dataset is employed for evaluation of the activity recognition classifier. Dataset was offered by Center for Advanced Studies in Adaptive Systems (CASAS) project, and was collected by monitoring a cognitively normal elderly resident by binary sensors for 21 months First, we converted the annotated binary sensor data into a binary activity images for corresponding activities. Then, activity images are used for training and testing the DCNN classifier. Finally, classifiers are evaluated with 10-fold cross validation method. Experimental results showed the best DCNN classifier gives 99.36% of accuracy. Our next step is to improve this classifier for detection of intertwined complex activities of elderly and to implement it on a real life long-term elderly monitoring system.