With the rapid deployment of Internet of Things (IoT) applications, video processing and streaming requirements are increasing, and edge computing is a good candidate to cope with strong latency and throughput expectations. In this paper, we consider the optimal routing, placement and scaling of IoT-based service function chains for object detection. We propose an edge networking approach dealing with limited CPU and network bandwidth resources in a joint optimization based on Integer Linear Programming for small problem instances, and a graph-based approximation to cope with scalability issues. We evaluate the efficiency of our algorithms through simulations and show that the graph-based approach converges towards a near-optimal solution in negligible time and is thus suitable for real-time function chain placement.