Some of the theoretical work is available with keywords like 'multi-object tracking' or 'multi-object detection' and also applied to tasks such as pedestrian tracking for autonomous vehicles, but there was also a bunch of applied work building and evaluating such CCTV systems which I can't find any links to now, sorry.
I didn't check if this research is still pursued or abandoned due to the rise of face recognition and other metric learning/re-identification technologies, but a few years ago a number of faculty at ETH Zurich were quite heavily involved in researching CCTV-based person tracking, which would work even against people taking measures against face or gait recognition algorithms if they weren't able to leave the CCTV-covered area. It was insinuated that various external state agencies, not all Switzerland-based, were specifically funding this research.
So, while the citizens might not want it, faculty seem perfectly happy to keep developing such technology for other clients.
The logic of this study is typical of many modern statistical studies:
1. the authors fit a certain model, adjusted for certain covariates, and observe that a certain effect of interest is statistically significant.
2. therefore, if (a) the model is "correct" i.e. sufficiently well approximates the observed data distribution, and (b) the regression coefficient and structural coefficient coincide i.e. the true generative DAG is compatible with the chosen set of covariates, then a causal association from the predictor to the outcome exists with high probability.
3. the authors simply assume (2) and therefore imply that a causal relation likely exists.
Direct action is an anarchist/autonomist idea of changing things directly rather than petitioning the government (see Graeber, Direct Action: an ethnography). It has literally nothing to do with fascist (particularly Nazi -- see Klemperer, Lingua Tertii Imperii) ideals of anti-intellectualism or action for action's sake beyond the word "action".