In this paper, we introduce a first of its kind, radio-signal based object detection system for controlled environments, which substitutes complex signal processing and expensive hardware with deep learning networks to detect patterns from low-quality, inexpensive sensors. Our system operates in the less crowded low-frequency range of 433 MHz in contrast to existing RF-based sensing methods and uses mini-Doppler maps generated from raw I/Q data, thereby allowing us to use cheap, off-the-shelf software defined radios. We demonstrate that our system is versatile enough to handle occlusions and is also sensitive to multiple objects; additionally, it does not use visual data and hence is not hampered by bad lighting. The core of our system is a VGG-16 based CNN architecture trained on the mini-Doppler maps. We achieve an accuracy of 0.96 on a binary classification task of detecting the presence or absence of an object in an enclosed space. Furthermore, we observe that our system shows promise for more complicated detection algorithms as it is able to successfully differentiate between the presence of a single object and two identical objects placed together. Our results indicate that convolutional networks can learn features important enough from spectrograms that enable it to distinguish the presence of objects, thereby eliminating the need of sophisticated signal processing methods to do the same.