wired up voice and vision, testing motion

r1/include/r1_voice/vision_context.hh

@@ -0,0 +1,37 @@
+#pragma once
+
+#include <string>
+
+namespace r1_voice {
+
+struct VisionConfig {
+    std::string detections_path =
+        "/dev/shm/r1_detections.json";
+
+    // Reject YOLO output older than this.
+    double max_age_seconds = 2.5;
+
+    double min_confidence = 0.35;
+};
+
+class VisionContext {
+public:
+    explicit VisionContext(
+        VisionConfig config
+    );
+
+    bool isVisionQuestion(
+        const std::string& text
+    ) const;
+
+    std::string buildContext() const;
+
+private:
+    VisionConfig config_;
+
+    static std::string lowerAscii(
+        std::string text
+    );
+};
+
+} // namespace r1_voice