With the progress of deep learning methods the last couple of years, object detection related tasks are improving rapidly. Using object detection for detecting guns in schools remove the need for human supervision and hopefully reduces police response time. This paper investigates how a gun detection system can be built by reading frames locally and using a server for detection. The detector is based on a pre-trained SSD model and through transfer learning is taught to recognize guns. The detector obtained an Average Precision of 51.1% and the server response time for a frame of size 1920 x 1080 was 480 ms, but could be scaled down to 240 x 135 to reach 210 ms, without affecting the accuracy. A non-gun class was implemented to reduce the number of false positives and on a set of 300 images containing 165 guns, the number of false positives dropped from 21 to 11.

This thesis consists of three chapters. The first chapter examines the impact of judicial discretion and left-digit bias on criminal sentencing outcomes. Judicial discretion allows judges to make nuanced decisions, taking into account details of legal cases that are not directly covered by law. However, judicial discretion can also expose behavioral biases and lead to irrational decision-making. I test for the existence of a particular behavioral bias: age-based left-digit bias. Specifically, I use a regression discontinuity design to test for changes in sentencing decisions occurring on an offender's 20th birthday using data on sentencing decisions from the state of Pennsylvania. I find that an offender sentenced just after his/her 20th birthday is 3.5 percentage points more likely to be sentenced to incarceration than an offender sentenced just before his/her 20th birthday. I test for evidence of conscious mechanisms underlying this effect and find no such evidence, leaving an unconscious bias as the best available explanation. Chapter two examines the impact of highly publicized police killings of black individuals on the racial gap in birth outcomes. Police killings of Black Americans are increasingly being met with significant media coverage and public response, including civil unrest. Given the frequency with which these events occur, it is vital to understand both their direct and indirect impacts. Using national birth certificate data and an event study design, I test for the impact of high-profile police-involved killings of Black Americans on racial disparities in maternal stress levels and birth outcomes. I find a large, statistically significant, and persistent increase in gestational hypertension of Black mothers relative to White mothers, strongly indicating an increase in the racial gap in maternal stress following these high-profile killings. I find limited evidence of an accompanying effect on the racial gap in birth outcomes. However, many existing papers similarly find no impacts of maternal stress on birth outcomes while simultaneously finding significant impacts on later-life outcomes, leaving room for additional future work based on these findings. How does the content of public communication by elected representatives change in response to highly salient, politically polarizing events? In Chapter 3, I examine this question using the text of tweets from members of the U.S. House of Representatives and the U.S. Senate, an n-gram text regression model and sentiment analysis alogorithms, and an event study design focused on mass shootings in the U.S. Observable effects on communication are concentrated on the day of and the day following a mass shooting. Republican members of Congress exhibit a reduced tweet frequency relative to Democratic members of Congress in the immediate aftermath of a shooting, while Democratic members of Congress speak with a more clearly differentiated Democratic vocabulary. Members from both parties speak with a more negative vocabulary. With Republicans collectively disengaging and Democrats collectively highlighting their partisan identification, this may suggest that Democrats are taking advantage of an opportunity for a political and/or policy win while Republicans in the same period are choosing to avoid additional political and/or policy losses.

Criminology / M. Tech. (Policing)

Criminology and Security Science / M. Tech. (Policing)

No description available.