Date of Award

June 2017

Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


Senem Velipasalar


Computer Vision, Fall Detection, Image Processing, Mobile Cameras, Step Counting, Traffic Sign Detection

Subject Categories



Wearable and mobile sensors have found widespread use in recent years due to their ever-decreasing cost, ease of deployment and use, and ability to provide continuous monitoring as opposed to sensors installed at fixed locations. Since many smart phones are now equipped with a variety of sensors, including accelerometer, gyroscope, magnetometer, microphone and camera, it has become more feasible to develop algorithms for activity monitoring, guidance and navigation of unmanned vehicles, autonomous driving and driver assistance, by using data from one or more of these sensors. In this thesis, we focus on multiple mobile camera applications, and present lightweight algorithms suitable for embedded mobile platforms. The mobile camera scenarios presented in the thesis are: (i) activity detection and step counting from wearable cameras, (ii) door detection for indoor navigation of unmanned vehicles, and (iii) traffic sign detection from vehicle-mounted cameras.

First, we present a fall detection and activity classification system developed for embedded smart camera platform CITRIC. In our system, the camera platform is worn by the subject, as opposed to static sensors installed at fixed locations in certain rooms, and, therefore, monitoring is not limited to confined areas, and extends to wherever the subject may travel including indoors and outdoors. Next, we present a real-time smart phone-based fall detection system, wherein we implement camera and accelerometer based fall-detection on Samsung Galaxy S™ 4. We fuse these two sensor modalities to have a more robust fall detection system. Then, we introduce a fall detection algorithm with autonomous thresholding using relative-entropy within the class of Ali-Silvey distance measures. As another wearable camera application, we present a footstep counting algorithm using a smart phone camera. This algorithm provides more accurate step-count compared to using only accelerometer data in smart phones and smart watches at various body locations.

As a second mobile camera scenario, we study autonomous indoor navigation of unmanned vehicles. A novel approach is proposed to autonomously detect and verify doorway openings by using the Google Project Tango™ platform.

The third mobile camera scenario involves vehicle-mounted cameras. More specifically, we focus on traffic sign detection from lower-resolution and noisy videos captured from vehicle-mounted cameras. We present a new method for accurate traffic sign detection, incorporating Aggregate Channel Features and Chain Code Histograms, with the goal of providing much faster training and testing, and comparable or better performance, with respect to deep neural network approaches, without requiring specialized processors. Proposed computer vision algorithms provide promising results for various useful applications despite the limited energy and processing capabilities of mobile devices.


Open Access

Included in

Engineering Commons